* [eval] increase timeout for swebench eval init/complete
* allow CmdRunAction to optionally block when .timeout is setted
* fix unit test for serialization
* fix unit tests for security analyzer
* fix integration tests
* add more timeout
* only check P2P when instances are non-empty;
convert P2P and F2P columns to string instead of list
---------
Co-authored-by: Graham Neubig <neubig@gmail.com>
* [eval] increase timeout for swebench eval init/complete
* allow CmdRunAction to optionally block when .timeout is setted
* fix unit test for serialization
* fix unit tests for security analyzer
* fix integration tests
* add more timeout
* Fail the Runtime tests check if the previous jobs were cancelled or failed
* Check only for failures
* Add test exit
* Test cancel
* fix conditions
* fix condition again
* Try another way
* Now test success
* move filematching logic into server
* wait until ready before returning
* show loading message instead of empty
* logspam
* delint
* fix type
* add a few more default ignores
* Reduce runtime tests duration by running them across CPUs
* fix hardcoded image name
* test two cpus
* Test folder change
* Up the CPU to 4 again to test
* Change to 3 CPUs
* Down to 2
* Add param to remove all openhands containers
* Add comment
* Add reruns just in case
* Fix ordering of if
* Add Handling of Cache Prompt When Formatting Messages
* Fix Value for Cache Control
* Fix Value for Cache Control
* Update openhands/core/message.py
Co-authored-by: Engel Nyst <enyst@users.noreply.github.com>
* Fix lint error
* Serialize Messages if Propt Caching Is Enabled
* Remove formatting message change
---------
Co-authored-by: Engel Nyst <enyst@users.noreply.github.com>
Co-authored-by: tobitege <10787084+tobitege@users.noreply.github.com>
* dummy test change
* regen yml: 1st install python 3.11, then poetry
* fix caching for poetry; old entry for python was rather useless
* fix steps order (cache before poetry)
* add poetry caching to ghcr_runtime; fix fork conditions
* ghcr_runtime: more caching actions; condition fixes
* fix interim action error (order of steps)
* cache@v4 instead of v3
* fixed interim typo for 2 fork conditions
* runtime/test_env_vars: compacted multiple tests into one to reduce time
* ugh if fork condition changes again
* (feat) making prompt caching optional instead of enabled default
At present, only the Claude models support prompt caching as a experimental feature, therefore, this feature should be implemented as an optional setting rather than being enabled by default.
Signed-off-by: Yi Lin <teroincn@gmail.com>
* handle the conflict
* fix unittest mock return value
* fix lint error in whitespace
---------
Signed-off-by: Yi Lin <teroincn@gmail.com>
* Update docs on LLM providers for consistency
* Update headless command
* minor tweaks based on feedback
---------
Co-authored-by: Robert Brennan <contact@rbren.io>
Co-authored-by: Robert Brennan <accounts@rbren.io>
* feat: add SWE-bench fullset support
* fix instance image list
* update eval script and documentation
* increase timeout for remote runtime
* add push script
* handle the case when ret push is an generator
* update pbar
* set SWE-Bench default to run SWE-Bench lite
* add script to cleanup remote runtime
* fix the cases when tag is too long
* update README
* update readme for cleanup
* rename od to oh
* Update evaluation/swe_bench/README.md
Co-authored-by: Graham Neubig <neubig@gmail.com>
* Update evaluation/swe_bench/README.md
Co-authored-by: Graham Neubig <neubig@gmail.com>
* Update evaluation/swe_bench/scripts/cleanup_remote_runtime.sh
Co-authored-by: Graham Neubig <neubig@gmail.com>
* Update evaluation/swe_bench/scripts/cleanup_remote_runtime.sh
Co-authored-by: Graham Neubig <neubig@gmail.com>
* Update evaluation/swe_bench/scripts/cleanup_remote_runtime.sh
Co-authored-by: Graham Neubig <neubig@gmail.com>
* gets API key and Runtime from env var
---------
Co-authored-by: Graham Neubig <neubig@gmail.com>
* update badges
* fix badges
* better badges
* move credits
* more badge work
* add gh logo
* update some copy
* update logo
* fix height
* update text
* emdash
* remove cruft
* move title
* update links
* add hr
* white logo
* move some stuff to getting-started
* revert logo
* more copy changes
* minor tweaks
* fix sidebar
* explicit sidebar
* words
* fix tag
* fix how-to
* more docs work
* update styles
* fix up custom sandbox docs
* change eval title
* fix up getting-started
* fix getting started
* update to 0.9.2
* update screenshot
* add company link
* fix dark mode
* minor fixes
* update image
* update headless and cli docs
* update readme
* fix links
* revert package
* rename links
* fix links
* fix link
* chagne to claude
* Add documentation for CLI mode
Fixes#3703
Add documentation for CLI mode in OpenHands.
* **New Documentation**: Add `docs/modules/usage/how-to/cli-mode.md` to document CLI mode.
- Include instructions on starting an interactive OpenHands session via the command line.
- Explain the difference between CLI mode and headless mode.
- Provide examples of CLI commands and expected outputs.
* **Update Existing Documentation**: Modify `docs/modules/usage/how-to/headless-mode.md`.
- Clarify the difference between headless mode and CLI mode.
- Add a reference to the new CLI mode documentation.
* Update cli-mode.md
* Update headless-mode.md
---------
Co-authored-by: Robert Brennan <accounts@rbren.io>
Co-authored-by: tofarr <tofarr@gmail.com>
* update badges
* fix badges
* better badges
* move credits
* more badge work
* add gh logo
* update some copy
* update logo
* fix height
* update text
* emdash
* remove cruft
* move title
* update links
* add hr
* white logo
* move some stuff to getting-started
* revert logo
* more copy changes
* minor tweaks
* words
* fix getting started
* update to 0.9.2
* update screenshot
* minor fixes
* CodeActAgent: fix message prep if prompt caching is not supported
* fix python version in regen tests workflow
* fix in conftest "mock_completion" method
* add disable_vision to LLMConfig; revert change in message parsing in llm.py
* format messages in several files for completion
* refactored message(s) formatting (llm.py); added vision_is_active()
* fix a unit test
* regenerate: added LOG_TO_FILE and FORCE_REGENERATE env flags
* try to fix path to logs folder in workflow
* llm: prevent index error
* try FORCE_USE_LLM in regenerate
* tweaks everywhere...
* fix 2 random unit test errors :(
* added FORCE_REGENERATE_TESTS=true to regenerate CLI
* fix test_lint_file_fail_typescript again
* double-quotes for env vars in workflow; llm logger set to debug
* fix typo in regenerate
* regenerate iterations now 20; applied iteration counter fix by Li
* regenerate: pass FORCE_REGENERATE flag into env
* fixes for int tests. several mock files updated.
* browsing_agent: fix response_parser.py adding ) to empty response
* test_browse_internet: fix skipif and revert obsolete mock files
* regenerate: fi bracketing for http server start/kill conditions
* disable test_browse_internet for CodeAct*Agents; mock files updated after merge
* missed to include more mock files earlier
* reverts after review feedback from Li
* forgot one
* browsing agent test, partial fixes and updated mock files
* test_browse_internet works in my WSL now!
* adapt unit test test_prompt_caching.py
* add DEBUG to regenerate workflow command
* convert regenerate workflow params to inputs
* more integration test mock files updated
* more files
* test_prompt_caching: restored test_prompt_caching_headers purpose
* file_ops: fix potential exception, like "cross device copy"; fixed mock files accordingly
* reverts/changes wrt feedback from xingyao
* updated docs and config template
* code cleanup wrt review feedback
* Catch exception and return finish action with an exception message in case of exception in llm completion
* Remove exception logs
* Raise llm response error for any exception in llm completion
* Raise LLMResponseError from async completion and async streaming completion as well
* feat: add SWE-bench fullset support
* fix instance image list
* update eval script and documentation
* increase timeout for remote runtime
* add push script
* handle the case when ret push is an generator
* update pbar
* set SWE-Bench default to run SWE-Bench lite
* feat: add SWE-bench fullset support
* fix instance image list
* update eval script and documentation
* add push script
* handle the case when ret push is an generator
* update pbar
Fix a potential issue that might lead to file corruption when edit linting is enabled
#3124 introduces a feature for editing: running linter twice before and after the change and only extract new errors introduced by the agent. This has some potential issues and I am working on #3649 to address them, but I feel like I am not gonna finish it in the next few days, and that PR has become harder and harder to review, thus this PR, which only focuses on a small improvement.
So what's the issue? When we run linters on the original file before our edits, we need to copy the original file and use a temporary file to lint, because linting may have side-effect (e.g. modifying the file in-place). I used the word "may" because:
Flake8 has no side-effect, so not a problem as of now.
We don't enforce this or document this "no side-effect" as a requirement for linter implementation, so side-effect is allowed.
Regardless, the "after-edit-linting" uses the same approach: backup the file before linting to avoid data corruption. We should keep our "before-edit-linting" consistent.
Why no new unittest that reproduces the issue? Well, as I have mentioned earlier, flake8 has no side-effect, so technically it's not a bug but a flaw. Therefore, there's no way to write a test that reproduces the issue.
* improve file editing prompts and unit test
converted most raise calls to a _output_error call in file_ops.py
* tweaks in test_agent_skill.py wrt to SEP separator
* tweaked the separator
* remove server runtime remnants and TEST_RUNTIME references
* restore use of TEST_RUNTIME args and variables
* fix integration tests
* added hint to properly escape docstrings
* revert latest prompt change
---------
Co-authored-by: Xingyao Wang <xingyao6@illinois.edu>
* change runtime to be synchronous
* fix test runtime with the new interface
* fix arg
* fix eval
* fix missing config attribute
* fix plugins
* fix on_event by revert it back to async
* update upload_file endpoint
* fix argument to upload file
* remove unncessary async for eval;
fix evaluation run in parallel
* use asyncio to run controller for eval
* revert file upload
* truncate eval test result output
* feat: refactor building logic into runtime builder
* return image name
* fix testcases
* use runtime builder for eventstream runtime
* have runtime builder return str
* add api_key to sandbox config
* draft remote runtime
* remove extra if clause
* initialize runtime based on box class
* add build logic
* use base64 for file upload
* get runtime image prefix from API
* replace ___ with _s_ to make it a valid image name
* use /build to start build and /build_status to check the build progress
* update logging
* fix exit code
* always use port
* add remote runtime
* rename runtime
* fix tests import
* make dir first if work_dir does not exists;
* update debug print to remote runtime
* fix exit close_sync
* update logging
* add retry for stop
* use all box class for test keep prompt
* fix test browsing
* add retry stop
* merge init commands to save startup time
* fix await
* remove sandbox url
* support execute through specific runtime url
* fix file ops
* simplify close
* factor out runtime retry code
* fix exception handling
* fix content type error (e.g., bad gateway when runtime is not ready)
* add retry for wait until alive;
add retry for check image exists
* Revert "add retry for wait until alive;"
This reverts commit dd013cd268.
* retry when wait until alive
* clean up msg
* directly save sdist to temp dir for _put_source_code_to_dir
* support running testcases in parallel
* tweak logging;
try to close session
* try to close session even on exception
* update poetry lock
* support remote to run integration tests
* add warning for workspace base on remote runtime
* set default runtime api
* remove server runtime
* update poetry lock
* support running swe-bench (n=1) eval on remoteruntime
* add a timeout of 30 min
* add todo for docker namespace
* update poetry loc
* linter and test_aider_linter extensions for eslint
* linter tweaks
* try enabling verbose output in linter test
* one more option for linter test
* try conftest.py for tests/unit folder
* enable verbose mode in workflow; remove conftest.py again
* debug print statements of linter results
* skip some tests if eslint is not installed at all
* more tweaks
* final test skip setups
* code quality revisions
* fix test again
---------
Co-authored-by: Graham Neubig <neubig@gmail.com>
* aider-bench: add visualization to summarize script and readme
* added example cost and actions histogram images for readme
* moved dependencies to evaluation section
* added optional START_ID env flag to resume from that instance id
* prepare_dataset: fix comparisons by using instance id's as int
* aider bench complete_runtime: close runtime to close container
* added matrix display of instance id for logging
* fix typo in summarize_results.py saying summarise_results
* changed start_id to skip_num to skip rows from dataset (start_id wasn't supportable)
* doc changes about huggingface spaces to temporarily point back to OD
* Add prompt caching
* remove anthropic-version from extra_headers
* change supports_prompt_caching method to attribute
* change caching strat and log cache statistics
* add reminder as a new message to fix caching
* fix unit test
* append reminder to the end of the last message content
* move token logs to post completion function
* fix unit test failure
* fix reminder and prompt caching
* unit tests for prompt caching
* add test
* clean up tests
* separate reminder, use latest two messages
* fix tests
---------
Co-authored-by: tobitege <10787084+tobitege@users.noreply.github.com>
Co-authored-by: Xingyao Wang <xingyao6@illinois.edu>
Co-authored-by: Engel Nyst <enyst@users.noreply.github.com>
* add eval_ids arg to run specific instance id's; fix/extend README
* fix description in parser for --eval-ids
* fix test_arg_parser.py to account for added arg
* fix typo in README to say "summarize" instead of "summarise" for script
* fix conftest.py option (#3573)
* try to fix fixture base_container_image in runtime conftest
* fix integration test mock files due to #3548
* fix test_ipython.py integration test
* Fix pypi build
The package on pypi only included opendevin/* (the poetry default). It also needs to include agenthub/*
* Bumped version so people will actually get it!
* Fix package definition
* Updated poetry lock file
* Update package name to openhands-ai
* Add py.typed to indicate that OpenHands has type annotations
* Replace package name with openhands_ai
* Fix tests to reflect new name
---------
Co-authored-by: Graham Neubig <neubig@gmail.com>
* Create helper functions
* Add map according to litellm docs
* Create ModelSelector
* Extend model selector
* use autocomplete from nextui
* Improve keys without providers
* Handle models without a provider
* Add verified section and some empty handling
* Add support for default or previously set models
* Update tests
* Lint
* Remove modifier
* Fix typescript error
* Functionality for switching to custom model
* Add verified models
* Respond to resetting to default
* Comment
* try to fix pip unavailable
* update test case for pip
* force rebuild in CI
* remove extra symlink
* fix newline
* added semi-colon to line 31
* Dockerfile.j2: activate env at the end
* Revert "Dockerfile.j2: activate env at the end"
This reverts commit cf2f565102.
* cleanup Dockerfile
* switch default python image
* remove image agnostic (no longer used)
* fix tests
* simplify integration tests default image
* add nodejs specific runtime tests
* update tests and workflows
* switch to nikolaik/python-nodejs:python3.11-nodejs22
* update build sh to output image name correctly
* increase custom images to test
* fix test
* fix test
* fix double quote
* try fixing ci
* update ghcr workflow
* fix artifact name
* try to fix ghcr again
* fix workflow
* save built image to correct dir
* remove extra -docker-image
* make last tag to be human readable image tag
* fix hyphen to underscore
* run test runtime on all tags
* revert app build
* separate ghcr workflow
* update dockerfile for eval
* fix tag for test run
* try fix tag
* try fix tag via matrix output
* try workflow again
* update comments
* try fixing test matrix
* fix artifact name
* try fix tag again
* Revert "try fix tag again"
This reverts commit b369badd8c.
* tweak filename
* try different path
* fix filepath
* try fix tag artifact path again
* save json instead of line
* update matrix
* print all tags in workflow
* support only streaming diff logs from the runtime client
* remove strip from log line to fix indentation
* get py interpreter for jupyter
* rstrip to remove newline on the rightside for logging
* fix blocking issue for stream logs
* set python interpreter path in bash ps1
* update testcase for jupyter py interpreter path
* remove accidentally added changes
* remove accidentally added changes
* only print dockerfile when debug
* add docs
* remove extra tests that weren't supposed to be in this pr
* add back missing test
* revert
* make LogBuffer synchronous to fix hang in integration tests
* fix integration tests
* Update opendevin/runtime/client/client.py
Co-authored-by: Engel Nyst <enyst@users.noreply.github.com>
* fix test case
* fix integration tests
* change deque to list
* update integration tests
* rename test runtime
* fix docs
* rename opendevin to openhands in tests
---------
Co-authored-by: tobitege <tobitege@gmx.de>
Co-authored-by: Graham Neubig <neubig@gmail.com>
Co-authored-by: tobitege <10787084+tobitege@users.noreply.github.com>
Co-authored-by: Engel Nyst <enyst@users.noreply.github.com>
* tweaks to linter.py to prep for typescript linting (not implemented yet)
* fix 2 linter unit tests
* simpler basic_lint output; updated unit test
* fix default gpt-4o model name in aider default config
* linter.py: use tsc (typescript compiler) for linting; added more tests
* make typescript linting be more forgiving
* use npx instead of npm to install typescript in Dockerfile.j2
* Fix merge mistake
* removed npx call from Dockerfile.j2
* fix run_cmd to use code parameter; replace regex in test
* fix test_lint_file_fail_typescript to ignore leading path characters
* added TODO comment to extract_error_line_from
* fixed bug in ts_lint with wrong line number parsing
* move test_runtime to ghcr_test_runtime; adapt workflows; fix runtime AttributeError
* split test_runtime.py into multiple tests in new tests/runtime folder
* moved common fixtures to tests/runtime/conftest.py
* add anthropic dependency for Vertex to Dockerfile.j2 template
* add anthropic also to pyproject
* revert Dockerfile.j2 change
Co-authored-by: Xingyao Wang <xingyao@all-hands.dev>
---------
Co-authored-by: Xingyao Wang <xingyao@all-hands.dev>
* renaming more opendevin occurences
* remove DOCKER_IMAGE variable from Makefile
* Revert rename in evaluation/swe_bench/run_infer.py
Co-authored-by: Xingyao Wang <xingyao@all-hands.dev>
---------
Co-authored-by: Xingyao Wang <xingyao@all-hands.dev>
* Fix: Reset iteration count when resuming due to throttling
* Fix inadvertent additions
* WIP
* Changing max_iterations instead of iteration count
* Now adjusting max_iterations or max_budget_per_task as appropriate
* Fix check on iterations
* Fix linter issues
* AgentController: remember initial max_iterations and use it to extend state's iterations
* increase task budget by initial value (not doubling it)
---------
Co-authored-by: Tim O'Farrell <tofarr@gmai.com>
Co-authored-by: tobitege <10787084+tobitege@users.noreply.github.com>
Co-authored-by: mamoodi <mamoodiha@gmail.com>
* try to fix pip unavailable
* update test case for pip
* force rebuild in CI
* remove extra symlink
* fix newline
* added semi-colon to line 31
* Dockerfile.j2: activate env at the end
* Revert "Dockerfile.j2: activate env at the end"
This reverts commit cf2f565102.
* cleanup Dockerfile
* switch default python image
* remove image agnostic (no longer used)
* fix tests
* simplify integration tests default image
* add nodejs specific runtime tests
* update tests and workflows
* switch to nikolaik/python-nodejs:python3.11-nodejs22
* update build sh to output image name correctly
* increase custom images to test
* fix test
* fix test
* fix double quote
* try fixing ci
* update ghcr workflow
* fix artifact name
* try to fix ghcr again
* fix workflow
* save built image to correct dir
* remove extra -docker-image
* make last tag to be human readable image tag
* fix hyphen to underscore
* run test runtime on all tags
* revert app build
* separate ghcr workflow
* update dockerfile for eval
* fix tag for test run
* try fix tag
* try fix tag via matrix output
* try workflow again
* update comments
* try fixing test matrix
* fix artifact name
* try fix tag again
* Revert "try fix tag again"
This reverts commit b369badd8c.
* tweak filename
* try different path
* fix filepath
* try fix tag artifact path again
* save json instead of line
* update matrix
* print all tags in workflow
* fix DOCKER_IMAGE to avoid ghcr.io/opendevin/ghcr.io/opendevin/od_runtime
* fix test matrix to only load unique test image tags
* try fix matrix again!!!!!
* add all runtime tests passed
---------
Co-authored-by: tobitege <tobitege@gmx.de>
Co-authored-by: Graham Neubig <neubig@gmail.com>
Co-authored-by: tobitege <10787084+tobitege@users.noreply.github.com>
* refactor agentskills to prepare for agentless
* fix import
* fix typo
* fix imports
* fix globals
* fix import
* fix import
* disable log to file to avoid auto-created log file w/ permission issue when import od in runtime
* import agentskills from OD instead from itself directly
* add back pythonpath
* remove chown since there's no log/folder
* add projects with licenses list
* fix projects
* fix projects
* fix projects
* fix licenses, include full Apache
* move AL 2.0 text to Credits
* Update README.md
Co-authored-by: Graham Neubig <neubig@gmail.com>
---------
Co-authored-by: Graham Neubig <neubig@gmail.com>
* Split Frontend and Python Unit tests
* Extra comment for deploy docs workflow
* Simpler comment
* Add paths and paths-ignore to unit tests
* More specific comment for py unit tests
* Remove paths-ignore because jobs are required
---------
Co-authored-by: Engel Nyst <enyst@users.noreply.github.com>
* fix potential flake8 miss checking
* Add unit test for edit_file_by_replace function with problematic file
* Add unit test for edit_file_by_replace function with problematic file
* Add unit test for edit_file_by_replace function with problematic file
* Add unit test for edit_file_by_replace function with problematic file
* Add unit test for edit_file function with problematic file
* Add unit test for edit_file function with problematic file
* Add unit test for edit_file function with problematic file
* Add unit test for edit_file function with problematic file
* Add unit test for edit_file function with problematic file
* Update opendevin/runtime/plugins/agent_skills/agentskills.py
Co-authored-by: Boxuan Li <liboxuan@connect.hku.hk>
* add test intention description
* fix potential flake8 miss checking
* fix potential flake8 miss checking
* fix potential flake8 miss checking
* fix potential flake8 miss checking
* fix potential flake8 miss checking
---------
Co-authored-by: Boxuan Li <liboxuan@connect.hku.hk>
Co-authored-by: Graham Neubig <neubig@gmail.com>
Co-authored-by: tobitege <tobitege@gmx.de>
* feat: Initial work on security analyzer
* feat: Add remote invariant client
* chore: improve fault tolerance of client
* feat: Add button to enable Invariant Security Analyzer
* [feat] confirmation mode for bash actions
* feat: Add Invariant Tab with security risk outputs
* feat: Add modal setting for Confirmation Mode
* fix: frontend tests for confirmation mode switch
* fix: add missing CONFIRMATION_MODE value in SettingsModal.test.tsx
* fix: update test to integrate new setting
* feat: Initial work on security analyzer
* feat: Add remote invariant client
* chore: improve fault tolerance of client
* feat: Add button to enable Invariant Security Analyzer
* feat: Add Invariant Tab with security risk outputs
* feat: integrate security analyzer with confirmation mode
* feat: improve invariant analyzer tab
* feat: Implement user confirmation for running bash/python code
* fix: don't display rejected actions
* fix: make confirmation show only on assistant messages
* feat: download traces, update policy, implement settings, auto-approve based on defined risk
* Fix: low risk not being shown because it's 0
* fix: duplicate logs in tab
* fix: log duplication
* chore: prepare for merge, remove logging
* Merge confirmation_mode from OpenDevin main
* test: update tests to pass
* chore: finish merging changes, security analyzer now operational again
* feat: document Security Analyzers
* refactor: api, monitor
* chore: lint, fix risk None, revert policy
* fix: check security_risk for None
* refactor: rename instances of invariant to security analyzer
* feat: add /api/options/security-analyzers endpoint
* Move security analyzer from tab to modal
* Temporary fix lock when security analyzer is not chosen
* feat: don't show lock at all when security analyzer is not enabled
* refactor:
- Frontend:
* change type of SECURITY_ANALYZER from bool to string
* add combobox to select SECURITY_ANALYZER, current options are "invariant and "" (no security analyzer)
* Security is now a modal, lock in bottom right is visible only if there's a security analyzer selected
- Backend:
* add close to SecurityAnalyzer
* instantiate SecurityAnalyzer based on provided string from frontend
* fix: update close to be async, to be consistent with other close on resources
* fix: max height of modal (prevent overflow)
* feat: add logo
* small fixes
* update docs for creating a security analyzer module
* fix linting
* update timeout for http client
* fix: move security_analyzer config from agent to session
* feat: add security_risk to browser actions
* add optional remark on combobox
* fix: asdict not called on dataclass, remove invariant dependency
* fix: exclude None values when serializing
* feat: take default policy from invariant-server instead of being hardcoded
* fix: check if policy is None
* update image name
* test: fix some failing runs
* fix: security analyzer tests
* refactor: merge confirmation_mode and security_analyzer into SecurityConfig. Change invariant error message for docker
* test: add tests for invariant parsing actions / observations
* fix: python linting for test_security.py
* Apply suggestions from code review
Co-authored-by: Engel Nyst <enyst@users.noreply.github.com>
* use ActionSecurityRisk | None intead of Optional
* refactor action parsing
* add extra check
* lint parser.py
* test: add field keep_prompt to test_security
* docs: add information about how to enable the analyzer
* test: Remove trailing whitespace in README.md text
---------
Co-authored-by: Mislav Balunovic <mislav.balunovic@gmail.com>
Co-authored-by: Engel Nyst <enyst@users.noreply.github.com>
Co-authored-by: Xingyao Wang <xingyao6@illinois.edu>
* remove unused plugin mixin
* change the entire jupyter PWD with bash;
print jupyter pwd in obs as well;
* remove unused field
* remove unused comments
* change the entire jupyter PWD with bash;
print jupyter pwd in obs as well;
* fix runtime tests for jupyter
* update intgeration tests
* fix test again
---------
Co-authored-by: Graham Neubig <neubig@gmail.com>
* try to fix pip unavailable
* update test case for pip
* force rebuild in CI
* remove extra symlink
* fix newline
* added semi-colon to line 31
* Dockerfile.j2: activate env at the end
* Revert "Dockerfile.j2: activate env at the end"
This reverts commit cf2f565102.
* cleanup Dockerfile
* switch default python image
* remove image agnostic (no longer used)
* fix tests
* switch to nikolaik/python-nodejs:python3.11-nodejs22
* fix test
* fix test
* revert docker
* update template
---------
Co-authored-by: tobitege <tobitege@gmx.de>
Co-authored-by: Graham Neubig <neubig@gmail.com>
* Fix issue where mouse drag fails on first attempt
* Detach event correctly
* Use old variable names
---------
Co-authored-by: Tim O'Farrell <tofarr@Tims-MacBook-Pro-2.local>
Co-authored-by: tobitege <tobitege@gmx.de>
* also copy over pyproject and poetry lock
* add missing readme
* remove extra git config init since it is already done in client.py
* only chown if the /workspace dir does not exists
* Revert "remove extra git config init since it is already done in client.py"
This reverts commit e8556cd76d.
* remove extra git config init since it is already done in client.py
* fix test runtime
* print container log while reconnecting
* print log in more readable format
* print log in more readable format
* increase lines
* clean up sandbox and ssh related stuff
* remove ssh hostname
* remove ssh hostname
* fix docker app cannot access runtime API issue
* remove ssh password
* API HOSTNAME should be pre-fixed with SANDBOX
* update config
* fix typo that breaks the test
* clean up sandbox and ssh related stuff
* remove ssh hostname
* remove ssh hostname
* remove ssh password
* update config
* fix typo that breaks the test
* Add unit tests for listen.py
* Added new tests
* Improve test coverage for listen.py
* Update tests
---------
Co-authored-by: opendevin <opendevin@all-hands.dev>
* switch default to eventstream runtime
* remove pull docker from makefile
* fix unittest
* fix file store path
* try deprecate server runtime
* remove persist sandbox
* move file utils
* remove server runtime related workflow
* remove unused method
* attempt to remove the reliance on filestore for BE
* fix async for list file
* fix list_files to post
* fix list files
* add suffix to directory
* make sure list file returns abs path;
make sure other backend endpoints accpets abs path
* remove server runtime test workflow
* set git config in runtime
* chown for workspace in client;
use INIT_COMMANDS to maintain all commands that need to be run before bash start;
* fix client issue;
add test case for git;
---------
Co-authored-by: Graham Neubig <neubig@gmail.com>
* switch default to eventstream runtime
* remove pull docker from makefile
* fix unittest
* fix file store path
* try deprecate server runtime
* remove persist sandbox
* move file utils
* remove server runtime related workflow
* remove unused method
* attempt to remove the reliance on filestore for BE
* fix async for list file
* fix list_files to post
* fix list files
* add suffix to directory
* make sure list file returns abs path;
make sure other backend endpoints accpets abs path
* remove server runtime test workflow
* set git config in runtime
* move multi-line bash tests to test_runtime;
support multi-line bash for esruntime;
* add testcase to handle PS2 prompt
* use bashlex for bash parsing to handle multi-line commands;
add testcases for multi-line commands
* revert ghcr runtime change
* Apply stash
* fix run as other user;
make test async;
* fix test runtime for run as od
* add run-as-devin to all the runtime tests
* handle the case when username is root
* move all run-as-devin tests from sandbox;
only tests a few cases on different user to save time;
* move over multi-line echo related tests to test_runtime
* fix user-specific jupyter by fixing the pypoetry virtualenv folder
* make plugin's init async;
chdir at initialization of jupyter plugin;
move ipy simple testcase to test runtime;
* support agentskills import in
move tests for jupyter pwd tests;
overload `add_env_vars` for EventStreamRuntime to update env var also in Jupyter;
make agentskills read env var lazily, in case env var is updated;
* fix ServerRuntime agentskills issue
* move agnostic image test to test_runtime
* merge runtime tests in CI
* fix enable auto lint as env var
* update warning message
* update warning message
* test for different container images
* change parsing output as debug
* add exception handling for update_pwd_decorator
* fix unit test indentation
* add plugins as default input to Runtime class;
remove init_sandbox_plugins;
implement add_env_var (include jupyter) in the base class;
* fix server runtime auto lint
* Revert "add exception handling for update_pwd_decorator"
This reverts commit 2b668b1506.
* tries to print debugging info for agentskills
* explictly setting uid (try fix permission issue)
* Revert "tries to print debugging info for agentskills"
This reverts commit 8be4c86756.
* set sandbox user id during testing to hopefully fix the permission issue
* add browser tools for server runtime
* try to debug for old pwd
* update debug cmd
* only test agnostic runtime when TEST_RUNTIME is Server
* fix temp dir mkdir
* load TEST_RUNTIME at the beginning
* remove ipython tests
* only log to file when DEBUG
* default logging to project root
* temporarily remove log to file
* fix LLM logger dir
* fix logger
* make set pwd an optional aux action
* fix prev pwd
* fix infinity recursion
* simplify
* do not import the whole od library to avoid logger folder by jupyter
* fix browsing
* increase timeout
* attempt to fix agentskills yet again
* clean up in testcases, since CI maybe run as non-root
* add _cause attribute for event.id
* remove parent
* add a bunch of debugging statement again for CI :(
* fix temp_dir fixture
* change all temp dir to follow pytest's tmp_path_factory
* remove extra bracket
* clean up error printing a bit
* jupyter chdir to self.config.workspace_mount_path_in_sandbox on initialization
* jupyter chdir to self.config.workspace_mount_path_in_sandbox on initialization
* add typing for tmp dir fixture
* clear the directory before running the test to avoid weird CI temp dir
* remove agnostic test case for server runtime
* Revert "remove agnostic test case for server runtime"
This reverts commit 30e2181c3f.
* disable agnostic tests in CI
* fix test
* make sure plugin arg is not passed when no plugin is specified;
remove redundant on_event function;
* move mock prompt
* rename runtime
* remove extra logging
* refactor run_controller's interface;
support multiple runtime for integration test;
filter out hostname for prompt
* uncomment other tests
* pass the right runtime to controller
* log runtime when start
* uncomment tests
* improve symbol filters
* add intergration test prompts that seemd ok
* add integration test workflow
* add python3 to default ubuntu image
* symlink python and fix permission to jupyter pip
* add retry for jupyter execute server
* fix jupyter pip install;
add post-process for jupyter pip install;
simplify init by add agent_skills path to PYTHONPATH;
add testcase to tests jupyter pip install;
* fix bug
* use ubuntu:22.04 for eventstream integration tests
* add todo
* update testcase
* remove redundant code
* fix unit test
* reduce dependency for runtime
* try making llama-index an optional dependency that's not installed by default
* remove pip install since it seemd not needed
* log ipython execution;
await write message since it returns a future
* update ipy testcase
* do not install llama-index in CI
* do not install llama-index in the app docker as well
* set sandbox container image in the integration test script
* log plugins & env var for runtime
* update conftest for sha256
* add git
* remove all non-alphanumeric chalracters
* add working ipy module tests!
* default to use host network
* remove is_async from browser to make thing a little more reliable;
retry loading browser when error;
* add sleep to wait a bit for http server
* kill http server before regenerate browsing tests
* fix browsing
* only set sandbox container image if undefined
* skip empty config value
* update evaluation to use the latest run_controller
* revert logger in execute_server to be compatible with server runtime
* revert logging level to fix jupyter
* set logger level
* revert the logging
* chmod for workspace to fix permission
* support getting timeout from action
* update test for server runtime
* try to fix file permission
* fix test_cmd_run_action_serialization_deserialization test (added timeout)
* poetry: pip 24.2, torch 2.2.2
* revert adding pip to pyproject.toml
* add build to dependencies in pyproject.toml
* forgot poetry lock --no-update
* fix a DelegatorAgent prompt_002.log (timeout)
* fix a DelegatorAgent prompt_003.log (timeout)
* couple more timeout attribs in prompt files
* some more prompt files
* prompts galore
* add clarification comment for timeout
* default timeout to config
* add assert
* update integraton tests for eventstream
* update integration tests
* fix timeout for action<->dict
* remove redundant on_event
* default to use instance image
* update run_controller interface
* add logging for copy
* refactor swe_bench for the new design
* fix action execution timeout
* updatelock
* remove build sandbox locally
* fix runtime
* use plain for-loop for single process
* remove extra print
* get swebench inference working
* print whole `test_result` dict
* got swebench patch post-process working
* update swe-bench evaluation readme
* refactor using shared reset_logger function
* move messy swebench prompt to a different file
* support the ability to specify whether to keep prompt
* support the ability to specify whether to keep prompt
* fix dockerfile
* fix import and remove unnecessary strip logic
* fix action serialization
* get agentbench running
* remove extra ls for agent bench
* fix agentbench metric
* factor out common documentation for eval
* update biocoder doc
* remove swe_env_box since it is no longer needed
* get biocoder working
* add func timeout for bird
* fix jupyter pwd with ~ as user name
* fix jupyter pwd with ~ as user name
* get bird working
* get browsing evaluation working
* make eda runnable
* fix id column
* fix eda run_infer
* unify eval output using a structured format;
make swebench coompatible with that format;
update client source code for every swebench run;
do not inject testcmd for swebench
* standardize existing benchs for the new eval output
* set update source code = true
* get gaia standardized
* fix gaia
* gorilla refactored but stuck at language.so to test
* refactor and make gpqa work
* refactor humanevalfix and get it working
* refactor logic reasoning and get it working
* refactor browser env so it works with eventstream runtime for eval
* add initial version of miniwob refactor
* fix browsergym environment
* get miniwob working!!
* allowing injecting additional dependency to OD runtime docker image
* allowing injecting additional dependency to OD runtime docker image
* support logic reasoning with pre-injected dependency
* get mint working
* update runtime build
* fix mint docker
* add test for keep_prompt;
add missing await close for some tests
* update integration tests for eventstream runtime
* fix integration tests for server runtime
* refactor ml bench and toolqa
* refactor webarena
* fix default factory
* Update run_infer.py
* add APIError to retry
* increase timeout for swebench
* make sure to hide api key when dump eval output
* update the behavior of put source code to put files instead of tarball
* add dishash to dependency
* sendintr when timeout
* fix dockerfile copy
* reduce timeout
* use dirhash to avoid repeat building for update source
* fix runtime_build testcase
* add dir_hash to docker build pipeline
* revert api error
* update poetry lock
* add retries for swebench run infer
* fix git patch
* update poetry lock
* adjust config order
* fix mount volumns
* enforce all eval to use "instance_id"
* remove file store from runtime
* make file_store public inside eventstream
* move the runtime logic inside `main` out
* support using async function for process_instance_fn
* refactor run_infer with the create_time
* fix file store
* Update evaluation/toolqa/utils.py
Co-authored-by: Graham Neubig <neubig@gmail.com>
* fix typo
---------
Co-authored-by: tobitege <tobitege@gmx.de>
Co-authored-by: super-dainiu <78588128+super-dainiu@users.noreply.github.com>
Co-authored-by: Graham Neubig <neubig@gmail.com>
* added workflow clean-up.yml to remove old workflows and artifacts
* dispatch-only run for now
* add retention-days of 14 to upload-artifact in ghcr.yml
* update the behavior of put source code to put files instead of tarball
* add dishash to dependency
* fix dockerfile copy
* use dirhash to avoid repeat building for update source
* fix runtime_build testcase
* add dir_hash to docker build pipeline
* add additional tests for source directory
* add comment
* clear the assertion by explictly check existing files
* also assert od is a dir
* ServerRuntime: config copy in init
* revert #3233 but more logging
* get_box_classes: reset order back to previous version
* 3 logging commands switched to debug (were info)
* runtimes debug output of config on initialization
* removed unneeded logger message from _init_container
* ghcr.yml with paths/paths-ignore conditions for pull_request
* include evaluation folder
* removed paths, just paths-ignore now
* deploy-docs only for docs
* exclude evaluation folder
Co-authored-by: Xingyao Wang <xingyao@all-hands.dev>
---------
Co-authored-by: Xingyao Wang <xingyao@all-hands.dev>
* support the ability to specify whether to keep prompt
* fix action serialization
* fix jupyter pwd with ~ as user name
* add test for keep_prompt;
add missing await close for some tests
* update integration tests for eventstream runtime
* fix integration tests for server runtime
* Remove global config from memory
* Remove runtime global config
* Remove from storage
* Remove global config
* Fix event stream tests
* Fix sandbox issue
* Change config
* Removed transferred tests
* Add swe env box
* Fixes on testing
* Fixed some tests
* Merge with stashed changes
* Fix typing
* Fix ipython test
* Revive function
* Make temp_dir fixture
* Remove test to avoid circular import
* fix eventstream filestore for test_runtime
* fix parse arg issue that cause integration test to fail
* support swebench pull from custom namespace
* add back simple tests for runtime
* move multi-line bash tests to test_runtime;
support multi-line bash for esruntime;
* add testcase to handle PS2 prompt
* use bashlex for bash parsing to handle multi-line commands;
add testcases for multi-line commands
* revert ghcr runtime change
* Apply stash
* fix run as other user;
make test async;
* fix test runtime for run as od
* add run-as-devin to all the runtime tests
* handle the case when username is root
* move all run-as-devin tests from sandbox;
only tests a few cases on different user to save time;
* move over multi-line echo related tests to test_runtime
* fix user-specific jupyter by fixing the pypoetry virtualenv folder
* make plugin's init async;
chdir at initialization of jupyter plugin;
move ipy simple testcase to test runtime;
* support agentskills import in
move tests for jupyter pwd tests;
overload `add_env_vars` for EventStreamRuntime to update env var also in Jupyter;
make agentskills read env var lazily, in case env var is updated;
* fix ServerRuntime agentskills issue
* move agnostic image test to test_runtime
* merge runtime tests in CI
* fix enable auto lint as env var
* update warning message
* update warning message
* test for different container images
* change parsing output as debug
* add exception handling for update_pwd_decorator
* fix unit test indentation
* add plugins as default input to Runtime class;
remove init_sandbox_plugins;
implement add_env_var (include jupyter) in the base class;
* fix server runtime auto lint
* Revert "add exception handling for update_pwd_decorator"
This reverts commit 2b668b1506.
* tries to print debugging info for agentskills
* explictly setting uid (try fix permission issue)
* Revert "tries to print debugging info for agentskills"
This reverts commit 8be4c86756.
* set sandbox user id during testing to hopefully fix the permission issue
* add browser tools for server runtime
* try to debug for old pwd
* update debug cmd
* only test agnostic runtime when TEST_RUNTIME is Server
* fix temp dir mkdir
* load TEST_RUNTIME at the beginning
* remove ipython tests
* only log to file when DEBUG
* default logging to project root
* temporarily remove log to file
* fix LLM logger dir
* fix logger
* make set pwd an optional aux action
* fix prev pwd
* fix infinity recursion
* simplify
* do not import the whole od library to avoid logger folder by jupyter
* fix browsing
* increase timeout
* attempt to fix agentskills yet again
* clean up in testcases, since CI maybe run as non-root
* add _cause attribute for event.id
* remove parent
* add a bunch of debugging statement again for CI :(
* fix temp_dir fixture
* change all temp dir to follow pytest's tmp_path_factory
* remove extra bracket
* clean up error printing a bit
* jupyter chdir to self.config.workspace_mount_path_in_sandbox on initialization
* jupyter chdir to self.config.workspace_mount_path_in_sandbox on initialization
* add typing for tmp dir fixture
* clear the directory before running the test to avoid weird CI temp dir
* remove agnostic test case for server runtime
* Revert "remove agnostic test case for server runtime"
This reverts commit 30e2181c3f.
* disable agnostic tests in CI
* fix test
* make sure plugin arg is not passed when no plugin is specified;
remove redundant on_event function;
* move mock prompt
* rename runtime
* remove extra logging
* refactor run_controller's interface;
support multiple runtime for integration test;
filter out hostname for prompt
* uncomment other tests
* pass the right runtime to controller
* log runtime when start
* uncomment tests
* improve symbol filters
* add intergration test prompts that seemd ok
* add integration test workflow
* add python3 to default ubuntu image
* symlink python and fix permission to jupyter pip
* add retry for jupyter execute server
* fix jupyter pip install;
add post-process for jupyter pip install;
simplify init by add agent_skills path to PYTHONPATH;
add testcase to tests jupyter pip install;
* fix bug
* use ubuntu:22.04 for eventstream integration tests
* add todo
* update testcase
* remove redundant code
* fix unit test
* reduce dependency for runtime
* try making llama-index an optional dependency that's not installed by default
* remove pip install since it seemd not needed
* log ipython execution;
await write message since it returns a future
* update ipy testcase
* do not install llama-index in CI
* do not install llama-index in the app docker as well
* set sandbox container image in the integration test script
* log plugins & env var for runtime
* update conftest for sha256
* add git
* remove all non-alphanumeric chalracters
* add working ipy module tests!
* default to use host network
* remove is_async from browser to make thing a little more reliable;
retry loading browser when error;
* add sleep to wait a bit for http server
* kill http server before regenerate browsing tests
* fix browsing
* only set sandbox container image if undefined
* skip empty config value
* update evaluation to use the latest run_controller
* revert logger in execute_server to be compatible with server runtime
* revert logging level to fix jupyter
* set logger level
* revert the logging
* chmod for workspace to fix permission
* support getting timeout from action
* update test for server runtime
* try to fix file permission
* fix test_cmd_run_action_serialization_deserialization test (added timeout)
* poetry: pip 24.2, torch 2.2.2
* revert adding pip to pyproject.toml
* add build to dependencies in pyproject.toml
* forgot poetry lock --no-update
* fix a DelegatorAgent prompt_002.log (timeout)
* fix a DelegatorAgent prompt_003.log (timeout)
* couple more timeout attribs in prompt files
* some more prompt files
* prompts galore
* add clarification comment for timeout
* default timeout to config
* add assert
* update integraton tests for eventstream
* update integration tests
* fix timeout for action<->dict
* remove redundant on_event
* fix action execution timeout
* updatelock
---------
Co-authored-by: Graham Neubig <neubig@gmail.com>
Co-authored-by: tobitege <tobitege@gmx.de>
* LLM class: added acompletion and streaming, unit test test_acompletion.py
* LLM: cleanup of self.config defaults and their use
* added set_missing_attributes to LLMConfig
* move default checker up
* add copy to
* implement for ServerRuntime
* implement copyto for runtime (required by eval);
add tests for copy to
* fix exist file check
* unify copy_to_behavior and fix stuff
* make sure codeact agent produce message in u/a/u/a order
* integration tests
* sync message changes to codeact swe
* fix integration tests
---------
Co-authored-by: Engel Nyst <enyst@users.noreply.github.com>
* reduce dependency for runtime
* try making llama-index an optional dependency that's not installed by default
* do not install llama-index in CI
* do not install llama-index in the app docker as well
* Remove global config from memory
* Remove runtime global config
* Remove from storage
* Remove global config
* Fix event stream tests
* Fix sandbox issue
* Change config
* Removed transferred tests
* Add swe env box
* Fixes on testing
* Fixed some tests
* Merge with stashed changes
* Fix typing
* Fix ipython test
* Revive function
* Make temp_dir fixture
* Remove test to avoid circular import
* fix eventstream filestore for test_runtime
* fix parse arg issue that cause integration test to fail
* support swebench pull from custom namespace
* add back simple tests for runtime
* move multi-line bash tests to test_runtime;
support multi-line bash for esruntime;
* add testcase to handle PS2 prompt
* use bashlex for bash parsing to handle multi-line commands;
add testcases for multi-line commands
* revert ghcr runtime change
* Apply stash
* fix run as other user;
make test async;
* fix test runtime for run as od
* add run-as-devin to all the runtime tests
* handle the case when username is root
* move all run-as-devin tests from sandbox;
only tests a few cases on different user to save time;
* move over multi-line echo related tests to test_runtime
* fix user-specific jupyter by fixing the pypoetry virtualenv folder
* make plugin's init async;
chdir at initialization of jupyter plugin;
move ipy simple testcase to test runtime;
* support agentskills import in
move tests for jupyter pwd tests;
overload `add_env_vars` for EventStreamRuntime to update env var also in Jupyter;
make agentskills read env var lazily, in case env var is updated;
* fix ServerRuntime agentskills issue
* move agnostic image test to test_runtime
* merge runtime tests in CI
* fix enable auto lint as env var
* update warning message
* update warning message
* test for different container images
* change parsing output as debug
* add exception handling for update_pwd_decorator
* fix unit test indentation
* add plugins as default input to Runtime class;
remove init_sandbox_plugins;
implement add_env_var (include jupyter) in the base class;
* fix server runtime auto lint
* Revert "add exception handling for update_pwd_decorator"
This reverts commit 2b668b1506.
* tries to print debugging info for agentskills
* explictly setting uid (try fix permission issue)
* Revert "tries to print debugging info for agentskills"
This reverts commit 8be4c86756.
* set sandbox user id during testing to hopefully fix the permission issue
* add browser tools for server runtime
* try to debug for old pwd
* update debug cmd
* only test agnostic runtime when TEST_RUNTIME is Server
* fix temp dir mkdir
* load TEST_RUNTIME at the beginning
* remove ipython tests
* only log to file when DEBUG
* default logging to project root
* temporarily remove log to file
* fix LLM logger dir
* fix logger
* make set pwd an optional aux action
* fix prev pwd
* fix infinity recursion
* simplify
* do not import the whole od library to avoid logger folder by jupyter
* fix browsing
* increase timeout
* attempt to fix agentskills yet again
* clean up in testcases, since CI maybe run as non-root
* add _cause attribute for event.id
* remove parent
* add a bunch of debugging statement again for CI :(
* fix temp_dir fixture
* change all temp dir to follow pytest's tmp_path_factory
* remove extra bracket
* clean up error printing a bit
* jupyter chdir to self.config.workspace_mount_path_in_sandbox on initialization
* jupyter chdir to self.config.workspace_mount_path_in_sandbox on initialization
* add typing for tmp dir fixture
* clear the directory before running the test to avoid weird CI temp dir
* remove agnostic test case for server runtime
* Revert "remove agnostic test case for server runtime"
This reverts commit 30e2181c3f.
* disable agnostic tests in CI
* fix test
---------
Co-authored-by: Graham Neubig <neubig@gmail.com>
* split_bash_commands replaced; temp_dir fixture fix in some tests
* tweak test_runtime
* skip 2 tests in test_runtime that need fixing in extra PR
* reverting bash parsing changes and re-enabled tests
* missed to revert a changed assert in test_runtime.py
* Remove global config from memory
* Remove runtime global config
* Remove from storage
* Remove global config
* Fix event stream tests
* Fix sandbox issue
* Change config
* Removed transferred tests
* Add swe env box
* Fixes on testing
* Fixed some tests
* Merge with stashed changes
* Fix typing
* Fix ipython test
* Revive function
* Make temp_dir fixture
* Remove test to avoid circular import
* fix eventstream filestore for test_runtime
* fix parse arg issue that cause integration test to fail
* support swebench pull from custom namespace
* add back simple tests for runtime
* move multi-line bash tests to test_runtime;
support multi-line bash for esruntime;
* add testcase to handle PS2 prompt
* use bashlex for bash parsing to handle multi-line commands;
add testcases for multi-line commands
* revert ghcr runtime change
---------
Co-authored-by: Graham Neubig <neubig@gmail.com>
I noticed that in dark modes the icons don't look as good as they could - making the outlines transparent makes them pop a bit more
Co-authored-by: Tim O'Farrell <tofarr@Tims-MacBook-Pro-2.local>
* colima: fix return code handling; added delay before retry; 4 retries
* moved docker context outside of function
* changed delete occurence; added logs output
* removed delete; trying to add more logging
* fix typo
* changed logging to github-style. maybe this finally shows up.
* reverted context; loop now with install+delete and alternating IP
* fix local keyword
* try limactl for creating an instance for IP
* revert IP change attempts
* actually return 0 in start_colima
* moved install out of loop again
* another to avoid duplicate start of colima via limactl
* added --init call for lima.yaml file creation
* dont trust an LLM to give you flags...
* Update run-unit-tests.yml
* colima: use a docker context specific to runner; prevent duplicate start
* updated use of context (for docker, not colima)
* added --ssh to colima start to use TCP instead of socket
* replace --ssh with random port
* test_runtime_client.py to test _execute_bash()
* runtime_build and runtime tweaks
* fix in docker script
* revert bash changes
* use sandbox_config.update_source_code to control source code update
* add od_version to the sandbox tag
* add doc instruction for update source code
* do not remove whole poetry folder;
add mamba clean
* add missing newlines
* cleanup runtime dockerfile into jinja template
* make prep temp file a separate function;
make that function accessible through cli
* modify `runtime_build.py` so it can generate directory for building docker img
* add dockerfile and sdist of runtime to gitignore since it will be dynamically generated
* add runtime to build
* do not rebuild new image when an `od_runtime` is provided
* use default container_image for testing if possible
* move runtime tests to ghcr runtime workflow
* update docker base dir for runtime
* fix unittest
* fix image name
* fix image name for test case
* rename to make it consistent
---------
Co-authored-by: tobitege <tobitege@gmx.de>
* fix: make max_budget_per_task optional in `run_agent_controller`
* update arg for each run infer
* fix: metrics logging carried along; reset llm metric with the agent;
---------
Co-authored-by: Graham Neubig <neubig@gmail.com>
* test_runtime_client.py to test _execute_bash()
* runtime_build and runtime tweaks
* fix in docker script
* revert bash changes
* use sandbox_config.update_source_code to control source code update
* add od_version to the sandbox tag
* add doc instruction for update source code
* do not remove whole poetry folder;
add mamba clean
* add missing newlines
---------
Co-authored-by: tobitege <tobitege@gmx.de>
* update and polish gptq eval
* fix typo
* Update evaluation/gpqa/README.md
Co-authored-by: Graham Neubig <neubig@gmail.com>
* Update evaluation/gpqa/run_infer.py
Co-authored-by: Graham Neubig <neubig@gmail.com>
* add headless mode to all appropriate agent controller call
* delegate set to error when in headless mode
* try to deduplicate a bit
* make headless_mode default to True and only change it to false for AgentSession
---------
Co-authored-by: Graham Neubig <neubig@gmail.com>
value:Thank you for taking the time to fill out this bug report. We greatly appreciate your effort to complete this template fully. Please provide as much information as possible to help us understand and address the issue effectively.
value:Thank you for taking the time to fill out this bug report. Please provide as much information as possible to help us understand and address the issue effectively.
- type:checkboxes
attributes:
label:Is there an existing issue for the same bug?
description:Please check if an issue already exists for the bug you encountered.
options:
- label:I have checked the troubleshooting document at https://docs.all-hands.dev/modules/usage/troubleshooting
required:true
- label:I have checked the existing issues.
required:true
- type:textarea
id:bug-description
attributes:
label:Describe the bug
description:Provide a short description of the problem.
label:Describe the bug and reproduction steps
description:Provide a description of the issue along with any reproduction steps.
validations:
required:true
- type:textarea
id:current-version
- type:dropdown
id:installation
attributes:
label:Current OpenDevin version
description:What version of OpenDevin are you using? If you're running in docker, tell us the tag you're using (e.g. ghcr.io/opendevin/opendevin:0.3.1).
render:bash
validations:
required:true
label:OpenHands Installation
description:How are you running OpenHands?
options:
- Docker command in README
- Development workflow
- app.all-hands.dev
- Other
default:0
- type:textarea
id:config
- type:input
id:openhands-version
attributes:
label:Installation and Configuration
description:Please provide any commands you ran and any configuration (redacting API keys)
render:bash
validations:
required:true
label:OpenHands Version
description:What version of OpenHands are you using?
placeholder:ex. 0.9.8, main, etc.
- type:textarea
id:model-agent
attributes:
label:Model and Agent
description:What model and agent are you using? You can see these settings in the UI by clicking the settings wheel.
placeholder:|
- Model:
- Agent:
- type:textarea
id:os-version
- type:dropdown
id:os
attributes:
label:Operating System
description:What Operating System are you using? Linux, Mac OS, WSL on Windows
- type:textarea
id:repro-steps
attributes:
label:Reproduction Steps
description:Please list the steps to reproduce the issue.
placeholder:|
1.
2.
3.
options:
- MacOS
- Linux
- WSL on Windows
- type:textarea
id:additional-context
attributes:
label:Logs, Errors, Screenshots, and Additional Context
description:If you want to share the chat history you can click the thumbs-down (👎) button above the input field and you will get a shareable link (you can also click thumbs up when things are going well of course!). LLM logs will be stored in the `logs/llm/default` folder. Please add any additional context about the problem here.
description:Please provide any additional information you think might help. If you want to share the chat history
you can click the thumbs-down (👎) button above the input field and you will get a shareable link
(you can also click thumbs up when things are going well of course!). LLM logs will be stored in the
`logs/llm/default` folder. Please add any additional context about the problem here.
poetry run ./evaluation/integration_tests/scripts/run_infer.sh llm.eval HEAD CodeActAgent '' $N_PROCESSES
# get evaluation report
REPORT_FILE=$(find evaluation/evaluation_outputs/outputs/integration_tests/CodeActAgent/deepseek-chat_maxiter_10_N* -name "report.md" -type f | head -n 1)
poetry run ./evaluation/swe_bench/scripts/run_infer.sh llm.eval HEAD CodeActAgent 300 30 $N_PROCESSES "princeton-nlp/SWE-bench_Lite" test
OUTPUT_FOLDER=$(find evaluation/evaluation_outputs/outputs/princeton-nlp__SWE-bench_Lite-test/CodeActAgent -name "deepseek-chat_maxiter_50_N_*-no-hint-run_1" -type d | head -n 1)
echo "OUTPUT_FOLDER for SWE-bench evaluation: $OUTPUT_FOLDER"
poetry run ./evaluation/swe_bench/scripts/eval_infer_remote.sh $OUTPUT_FOLDER/output.jsonl $N_PROCESSES "princeton-nlp/SWE-bench_Lite" test
poetry run ./evaluation/swe_bench/scripts/eval/summarize_outputs.py $OUTPUT_FOLDER/output.jsonl > summarize_outputs.log 2>&1
echo "SWEBENCH_REPORT<<EOF" >> $GITHUB_ENV
cat summarize_outputs.log >> $GITHUB_ENV
echo "EOF" >> $GITHUB_ENV
- name:Create tar.gz of evaluation outputs
run:|
TIMESTAMP=$(date +'%y-%m-%d-%H-%M')
tar -czvf evaluation_outputs_${TIMESTAMP}.tar.gz evaluation/evaluation_outputs/outputs
# Workflow that marks issues and PRs with no activity for 30 days with "Stale" and closes them after 7 more days of no activity
name:'Close stale issues'
# Runs every day at 01:30
on:
schedule:
- cron:'30 1 * * *'
@@ -9,21 +12,10 @@ jobs:
steps:
- uses:actions/stale@v9
with:
# Aggressively close issues that have been explicitly labeled `age-out`
any-of-labels:age-out
stale-issue-message:'This issue is stale because it has been open for 7 days with no activity. Remove stale label or comment or this will be closed in 1 day.'
close-issue-message:'This issue was closed because it has been stalled for over 7 days with no activity.'
stale-pr-message:'This PR is stale because it has been open for 7 days with no activity. Remove stale label or comment or this will be closed in 1 days.'
close-pr-message:'This PR was closed because it has been stalled for over 7 days with no activity.'
days-before-stale:7
days-before-close:1
- uses:actions/stale@v9
with:
# Be more lenient with other issues
stale-issue-message:'This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.'
close-issue-message:'This issue was closed because it has been stalled for over 30 days with no activity.'
stale-pr-message:'This PR is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.'
close-pr-message:'This PR was closed because it has been stalled for over 30 days with no activity.'
days-before-stale:30
exempt-issue-labels:'tracked'
close-issue-message:'This issue was closed because it has been stalled for over 30 days with no activity.'
close-pr-message:'This PR was closed because it has been stalled for over 30 days with no activity.'
OpenHands is an automated AI software engineer. It is a repo with a Python backend
(in the `openhands` directory) and TypeScript frontend (in the `frontend` directory).
General Setup:
- To set up the entire repo, including frontend and backend, run `make build`
- To run linting and type-checking before finishing the job, run `poetry run pre-commit run --all-files --config ./dev_config/python/.pre-commit-config.yaml`
Backend:
- Located in the `openhands` directory
- Testing:
- All tests are in `tests/unit/test_*.py`
- To test new code, run `poetry run pytest tests/unit/test_xxx.py` where `xxx` is the appropriate file for the current functionality
- Write all tests with pytest
Frontend:
- Located in the `frontend` directory
- Prerequisites: A recent version of NodeJS / NPM
- Setup: Run `npm install` in the frontend directory
- Testing:
- Run tests: `npm run test`
- To run specific tests: `npm run test -- -t "TestName"`
- Building:
- Build for production: `npm run build`
- Environment Variables:
- Set in `frontend/.env` or as environment variables
- Available variables: VITE_BACKEND_HOST, VITE_USE_TLS, VITE_INSECURE_SKIP_VERIFY, VITE_FRONTEND_PORT
- Internationalization:
- Generate i18n declaration file: `npm run make-i18n`
Thanks for your interest in contributing to OpenDevin! We welcome and appreciate contributions.
Thanks for your interest in contributing to OpenHands! We welcome and appreciate contributions.
## How Can I Contribute?
There are many ways that you can contribute:
1.**Download and use** OpenDevin, and send [issues](https://github.com/OpenDevin/OpenDevin/issues) when you encounter something that isn't working or a feature that you'd like to see.
2.**Send feedback** after each session by [clicking the thumbs-up thumbs-down buttons](https://docs.all-hands.dev/modules/usage/feedback), so we can see where things are working and failing, and also build an open dataset for training code agents.
3.**Improve the Codebase** by sending PRs (see details below). In particular, we have some [good first issue](https://github.com/OpenDevin/OpenDevin/labels/good%20first%20issue) issues that may be ones to start on.
## Understanding OpenDevin's CodeBase
## Understanding OpenHands's CodeBase
To understand the codebase, please refer to the README in each module:
- [frontend](./frontend/README.md)
- [agenthub](./agenthub/README.md)
- [evaluation](./evaluation/README.md)
- [opendevin](./opendevin/README.md)
- [server](./opendevin/server/README.md)
- [openhands](./openhands/README.md)
- [agenthub](./openhands/agenthub/README.md)
- [server](./openhands/server/README.md)
## Setting up your development environment
We have a separate doc [Development.md](https://github.com/All-Hands-AI/OpenHands/blob/main/Development.md) that tells you how to set up a development workflow.
## How can I contribute?
There are many ways that you can contribute:
1.**Download and use** OpenHands, and send [issues](https://github.com/All-Hands-AI/OpenHands/issues) when you encounter something that isn't working or a feature that you'd like to see.
2.**Send feedback** after each session by [clicking the thumbs-up thumbs-down buttons](https://docs.all-hands.dev/modules/usage/feedback), so we can see where things are working and failing, and also build an open dataset for training code agents.
3.**Improve the Codebase** by sending PRs (see details below). In particular, we have some [good first issues](https://github.com/All-Hands-AI/OpenHands/labels/good%20first%20issue) that may be ones to start on.
## What can I build?
Here are a few ways you can help improve the codebase.
#### UI/UX
We're always looking to improve the look and feel of the application. If you've got a small fix
for something that's bugging you, feel free to open up a PR that changes the `./frontend` directory.
If you're looking to make a bigger change, add a new UI element, or significantly alter the style
of the application, please open an issue first, or better, join the #frontend channel in our Slack
to gather consensus from our design team first.
#### Improving the agent
Our main agent is the CodeAct agent. You can [see its prompts here](https://github.com/All-Hands-AI/OpenHands/tree/main/openhands/agenthub/codeact_agent)
Changes to these prompts, and to the underlying behavior in Python, can have a huge impact on user experience.
You can try modifying the prompts to see how they change the behavior of the agent as you use the app
locally, but we will need to do an end-to-end evaluation of any changes here to ensure that the agent
is getting better over time.
We use the [SWE-bench](https://www.swebench.com/) benchmark to test our agent. You can join the #evaluation
channel in Slack to learn more.
#### Adding a new agent
You may want to experiment with building new types of agents. You can add an agent to `openhands/agenthub`
to help expand the capabilities of OpenHands.
#### Adding a new runtime
The agent needs a place to run code and commands. When you run OpenHands on your laptop, it uses a Docker container
to do this by default. But there are other ways of creating a sandbox for the agent.
If you work for a company that provides a cloud-based runtime, you could help us add support for that runtime
by implementing the [interface specified here](https://github.com/All-Hands-AI/OpenHands/blob/main/openhands/runtime/runtime.py).
#### Testing
When you write code, it is also good to write tests. Please navigate to the `tests` folder to see existing test suites.
At the moment, we have two kinds of tests: `unit` and `integration`. Please refer to the README for each test suite. These tests also run on GitHub's continuous integration to ensure quality of the project.
## Sending Pull Requests to OpenDevin
## Sending Pull Requests to OpenHands
### 1. Fork the Official Repository
Fork the [OpenDevin repository](https://github.com/OpenDevin/OpenDevin) into your own account.
Clone your own forked repository into your local environment:
You'll need to fork our repository to send us a Pull Request. You can learn more
about how to fork a GitHub repo and open a PR with your changes in [this article](https://medium.com/swlh/forks-and-pull-requests-how-to-contribute-to-github-repos-8843fac34ce8)
Set the official repository as your [upstream](https://www.atlassian.com/git/tutorials/git-forks-and-upstreams) to synchronize with the latest update in the official repository.
You should see both `origin` and `upstream` in the output.
### 3. Synchronize with Official Repository
Synchronize latest commit with official repository before coding:
```shell
git fetch upstream
git checkout main
git merge upstream/main
git push origin main
```
### 4. Set up the Development Environment
We have a separate doc [Development.md](https://github.com/OpenDevin/OpenDevin/blob/main/Development.md) that tells you how to set up a development workflow.
### 5. Write Code and Commit It
Once you have done this, you can write code, test it, and commit it to a branch (replace `my_branch` with an appropriate name):
```shell
git checkout -b my_branch
git add .
git commit
git push origin my_branch
```
### 6. Open a Pull Request
* On GitHub, go to the page of your forked repository, and create a Pull Request:
- Click on `Branches`
- Click on the `...` beside your branch and click on `New pull request`
- Set `base repository` to `OpenDevin/OpenDevin`
- Set `base` to `main`
- Click `Create pull request`
The PR should appear in [OpenDevin PRs](https://github.com/OpenDevin/OpenDevin/pulls).
Then the OpenDevin team will review your code.
## PR Rules
### 1. Pull Request title
### Pull Request title
As described [here](https://github.com/commitizen/conventional-commit-types/blob/master/index.json), a valid PR title should begin with one of the following prefixes:
-`feat`: A new feature
@@ -109,8 +84,11 @@ For example, a PR title could be:
-`refactor: modify package path`
-`feat(frontend): xxxx`, where `(frontend)` means that this PR mainly focuses on the frontend component.
You may also check out previous PRs in the [PR list](https://github.com/OpenDevin/OpenDevin/pulls).
You may also check out previous PRs in the [PR list](https://github.com/All-Hands-AI/OpenHands/pulls).
### 2. Pull Request description
### Pull Request description
- If your PR is small (such as a typo fix), you can go brief.
- If it contains a lot of changes, it's better to write more details.
If your changes are user-facing (e.g. a new feature in the UI, a change in behavior, or a bugfix)
please include a short message that we can add to our changelog.
We would like to thank all the [contributors](https://github.com/All-Hands-AI/OpenHands/graphs/contributors) who have helped make OpenHands possible. We greatly appreciate your dedication and hard work.
## Open Source Projects
OpenHands includes and adapts the following open source projects. We are grateful for their contributions to the open source community:
- Description: AI pair programming tool. OpenHands has adapted and integrated its linter module for code-related tasks in [`agentskills utilities`](https://github.com/All-Hands-AI/OpenHands/tree/main/openhands/runtime/plugins/agent_skills/utils/aider)
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
### BSD 3-Clause License
Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
3. Neither the name of the copyright holder nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
### Apache License 2.0
Apache License
Version 2.0, January 2004
@@ -189,14 +268,45 @@
Copyright [yyyy] [name of copyright owner]
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
This guide is for people working on OpenDevin and editing the source code.
If you wish to contribute your changes, check out the [CONTRIBUTING.md](https://github.com/OpenDevin/OpenDevin/blob/main/CONTRIBUTING.md) on how to clone and setup the project initially before moving on.
Otherwise, you can clone the OpenDevin project directly.
This guide is for people working on OpenHands and editing the source code.
If you wish to contribute your changes, check out the [CONTRIBUTING.md](https://github.com/All-Hands-AI/OpenHands/blob/main/CONTRIBUTING.md) on how to clone and setup the project initially before moving on.
Otherwise, you can clone the OpenHands project directly.
## Start the server for development
### 1. Requirements
* Linux, Mac OS, or [WSL on Windows](https://learn.microsoft.com/en-us/windows/wsl/install) [Ubuntu <= 22.04]
* Linux, Mac OS, or [WSL on Windows](https://learn.microsoft.com/en-us/windows/wsl/install) [Ubuntu <= 22.04]
* [Docker](https://docs.docker.com/engine/install/) (For those on MacOS, make sure to allow the default Docker socket to be used from advanced settings!)
Begin by building the project which includes setting up the environment and installing dependencies. This step ensures that OpenDevin is ready to run on your system:
Begin by building the project which includes setting up the environment and installing dependencies. This step ensures that OpenHands is ready to run on your system:
```bash
make build
```
### 3. Configuring the Language Model
OpenDevin supports a diverse array of Language Models (LMs) through the powerful [litellm](https://docs.litellm.ai) library. By default, we've chosen the mighty GPT-4 from OpenAI as our go-to model, but the world is your oyster! You can unleash the potential of Anthropic's suave Claude, the enigmatic Llama, or any other LM that piques your interest.
OpenHands supports a diverse array of Language Models (LMs) through the powerful [litellm](https://docs.litellm.ai) library. By default, we've chosen the mighty GPT-4 from OpenAI as our go-to model, but the world is your oyster! You can unleash the potential of Anthropic's suave Claude, the enigmatic Llama, or any other LM that piques your interest.
To configure the LM of your choice, run:
@@ -44,20 +46,20 @@ To configure the LM of your choice, run:
make setup-config
```
This command will prompt you to enter the LLM API key, model name, and other variables ensuring that OpenDevin is tailored to your specific needs. Note that the model name will apply only when you run headless. If you use the UI, please set the model in the UI.
This command will prompt you to enter the LLM API key, model name, and other variables ensuring that OpenHands is tailored to your specific needs. Note that the model name will apply only when you run headless. If you use the UI, please set the model in the UI.
Note: If you have previously run OpenDevin using the docker command, you may have already set some environmental variables in your terminal. The final configurations are set from highest to lowest priority:
Note: If you have previously run OpenHands using the docker command, you may have already set some environmental variables in your terminal. The final configurations are set from highest to lowest priority:
Some alternative models may prove more challenging to tame than others. Fear not, brave adventurer! We shall soon unveil LLM-specific documentation to guide you on your quest.
And if you've already mastered the art of wielding a model other than OpenAI's GPT, we encourage you to share your setup instructions with us by creating instructions and adding it [to our documentation](https://github.com/OpenDevin/OpenDevin/tree/main/docs/modules/usage/llms).
And if you've already mastered the art of wielding a model other than OpenAI's GPT, we encourage you to share your setup instructions with us by creating instructions and adding it [to our documentation](https://github.com/All-Hands-AI/OpenHands/tree/main/docs/modules/usage/llms).
For a full list of the LM providers and models available, please consult the [litellm documentation](https://docs.litellm.ai/docs/providers).
### 4. Running the application
#### Option A: Run the Full Application
Once the setup is complete, launching OpenDevin is as simple as running a single command. This command starts both the backend and frontend servers seamlessly, allowing you to interact with OpenDevin:
Once the setup is complete, launching OpenHands is as simple as running a single command. This command starts both the backend and frontend servers seamlessly, allowing you to interact with OpenHands:
```bash
make run
```
@@ -75,10 +77,10 @@ make run
### 6. LLM Debugging
If you encounter any issues with the Language Model (LM) or you're simply curious, you can inspect the actual LLM prompts and responses. To do so, export DEBUG=1 in the environment and restart the backend.
OpenDevin will then log the prompts and responses in the logs/llm/CURRENT_DATE directory, allowing you to identify the causes.
OpenHands will then log the prompts and responses in the logs/llm/CURRENT_DATE directory, allowing you to identify the causes.
### 7. Help
Need assistance or information on available targets and commands? The help command provides all the necessary guidance to ensure a smooth experience with OpenDevin.
Need assistance or information on available targets and commands? The help command provides all the necessary guidance to ensure a smooth experience with OpenHands.
```bash
make help
```
@@ -91,9 +93,36 @@ To run tests, refer to the following:
poetry run pytest ./tests/unit/test_*.py
```
#### Integration tests
Please refer to [this README](./tests/integration/README.md) for details.
### 9. Add or update dependency
1. Add your dependency in `pyproject.toml` or use `poetry add xxx`
2. Update the poetry.lock file via `poetry lock --no-update`
### 9. Use existing Docker image
To reduce build time (e.g., if no changes were made to the client-runtime component), you can use an existing Docker container image. Follow these steps:
1. Set the SANDBOX_RUNTIME_CONTAINER_IMAGE environment variable to the desired Docker image.
<a href="https://docs.all-hands.dev/modules/usage/getting-started"><img src="https://img.shields.io/badge/Documentation-000?logo=googledocs&logoColor=FFE165&style=for-the-badge" alt="Check out the documentation"></a>
<a href="https://arxiv.org/abs/2407.16741"><img src="https://img.shields.io/badge/Paper%20on%20Arxiv-000?logoColor=FFE165&logo=arxiv&style=for-the-badge" alt="Paper on Arxiv"></a>
<h1 align="center">OpenDevin: Code Less, Make More</h1>
<a href="https://docs.all-hands.dev/modules/usage/intro"><img src="https://img.shields.io/badge/Documentation-OpenDevin-blue?logo=googledocs&logoColor=white&style=for-the-badge" alt="Check out the documentation"></a>
> By default, this command pulls the `latest` tag, which represents the most recent release of OpenDevin. You have other options as well:
> - For a specific release version, use `ghcr.io/opendevin/opendevin:<OpenDevin_version>` (replace <OpenDevin_version> with the desired version number).
> - For the most up-to-date development version, use `ghcr.io/opendevin/opendevin:main`. This version may be **(unstable!)** and is recommended for testing or development purposes only.
>
> Choose the tag that best suits your needs based on stability requirements and desired features.
You'll find OpenHands running at [http://localhost:3000](http://localhost:3000)!
You'll find OpenDevin running at [http://localhost:3000](http://localhost:3000) with access to `./workspace`. To have OpenDevin operate on your code, place it in `./workspace`.
OpenDevin will only have access to this workspace folder. The rest of your system will not be affected as it runs in a secured docker sandbox.
Finally, you'll need a model provider and API key.
[Anthropic's Claude 3.5 Sonnet](https://www.anthropic.com/api) (`anthropic/claude-3-5-sonnet-20241022`)
works best, but you have [many options](https://docs.all-hands.dev/modules/usage/llms).
Upon opening OpenDevin, you must select the appropriate `Model` and enter the `API Key` within the settings that should pop up automatically. These can be set at any time by selecting
the `Settings` button (gear icon) in the UI. If the required `Model` does not exist in the list, you can manually enter it in the text box.
---
For the development workflow, see [Development.md](https://github.com/OpenDevin/OpenDevin/blob/main/Development.md).
You can also [connect OpenHands to your local filesystem](https://docs.all-hands.dev/modules/usage/runtimes),
run OpenHands in a scriptable [headless mode](https://docs.all-hands.dev/modules/usage/how-to/headless-mode),
interact with it via a [friendly CLI](https://docs.all-hands.dev/modules/usage/how-to/cli-mode),
or run it on tagged issues with [a github action](https://github.com/All-Hands-AI/OpenHands-resolver).
Are you having trouble? Check out our [Troubleshooting Guide](https://docs.all-hands.dev/modules/usage/troubleshooting).
Visit [Installation](https://docs.all-hands.dev/modules/usage/installation) for more information and setup instructions.
## 🚀 Documentation
If you want to modify the OpenHands source code, check out [Development.md](https://github.com/All-Hands-AI/OpenHands/blob/main/Development.md).
To learn more about the project, and for tips on using OpenDevin,
**check out our [documentation](https://docs.all-hands.dev/modules/usage/intro)**.
Having issues? The [Troubleshooting Guide](https://docs.all-hands.dev/modules/usage/troubleshooting) can help.
There you'll find resources on how to use different LLM providers (like ollama and Anthropic's Claude),
## 📖 Documentation
To learn more about the project, and for tips on using OpenHands,
**check out our [documentation](https://docs.all-hands.dev/modules/usage/getting-started)**.
There you'll find resources on how to use different LLM providers,
troubleshooting resources, and advanced configuration options.
## 🤝 How to Contribute
OpenDevin is a community-driven project, and we welcome contributions from everyone.
OpenHands is a community-driven project, and we welcome contributions from everyone.
Whether you're a developer, a researcher, or simply enthusiastic about advancing the field of
software engineering with AI, there are many ways to get involved:
- **Code Contributions:** Help us develop new agents, core functionality, the frontend and other interfaces, or sandboxing solutions.
- **Research and Evaluation:** Contribute to our understanding of LLMs in software engineering, participate in evaluating the models, or suggest improvements.
- **Feedback and Testing:** Use the OpenDevin toolset, report bugs, suggest features, or provide feedback on usability.
- **Feedback and Testing:** Use the OpenHands toolset, report bugs, suggest features, or provide feedback on usability.
For details, please check [CONTRIBUTING.md](./CONTRIBUTING.md).
## 🤖 Join Our Community
Whether you're a developer, a researcher, or simply enthusiastic about OpenDevin, we'd love to have you in our community.
Whether you're a developer, a researcher, or simply enthusiastic about OpenHands, we'd love to have you in our community.
Let's make software engineering better together!
- [Slack workspace](https://join.slack.com/t/opendevin/shared_invite/zt-2jsrl32uf-fTeeFjNyNYxqSZt5NPY3fA) - Here we talk about research, architecture, and future development.
- [Slack workspace](https://join.slack.com/t/openhands-ai/shared_invite/zt-2tom0er4l-JeNUGHt_AxpEfIBstbLPiw) - Here we talk about research, architecture, and future development.
- [Discord server](https://discord.gg/ESHStjSjD4) - This is a community-run server for general discussion, questions, and feedback.
OpenHands is built by a large number of contributors, and every contribution is greatly appreciated! We also build upon other open source projects, and we are deeply thankful fortheir work.
For a list of open source projects and licenses used in OpenHands, please see our [CREDITS.md](./CREDITS.md) file.
## 📚 Cite
```
@misc{opendevin2024,
author = {{OpenDevin Team}},
title = {{OpenDevin: An Open Platform for AI Software Developers as Generalist Agents}},
note = {Accessed: ENTER THE DATE YOU ACCESSED THE PROJECT}
@misc{openhands,
title={{OpenHands: An Open Platform for AI Software Developers as Generalist Agents}},
author={Xingyao Wang and Boxuan Li and Yufan Song and Frank F. Xu and Xiangru Tang and Mingchen Zhuge and Jiayi Pan and Yueqi Song and Bowen Li and Jaskirat Singh and Hoang H. Tran and Fuqiang Li and Ren Ma and Mingzhang Zheng and Bill Qian and Yanjun Shao and Niklas Muennighoff and Yizhe Zhang and Binyuan Hui and Junyang Lin and Robert Brennan and Hao Peng and Heng Ji and Graham Neubig},
In this folder, there may exist multiple implementations of `Agent` that will be used by the framework.
For example, `agenthub/codeact_agent`, etc.
Contributors from different backgrounds and interests can choose to contribute to any (or all!) of these directions.
## Constructing an Agent
The abstraction for an agent can be found [here](../opendevin/controller/agent.py).
Agents are run inside of a loop. At each iteration, `agent.step()` is called with a
[State](../opendevin/controller/state/state.py) input, and the agent must output an [Action](../opendevin/events/action).
Every agent also has a `self.llm` which it can use to interact with the LLM configured by the user.
See the [LiteLLM docs for `self.llm.completion`](https://docs.litellm.ai/docs/completion).
## State
The `state` contains:
- A history of actions taken by the agent, as well as any observations (e.g. file content, command output) from those actions
- A list of actions/observations that have happened since the most recent step
- A [`root_task`](https://github.com/OpenDevin/OpenDevin/blob/main/opendevin/controller/state/task.py), which contains a plan of action
- The agent can add and modify subtasks through the `AddTaskAction` and `ModifyTaskAction`
## Actions
Here is a list of available Actions, which can be returned by `agent.step()`:
- [`CmdRunAction`](../opendevin/events/action/commands.py) - Runs a command inside a sandboxed terminal
- [`IPythonRunCellAction`](../opendevin/events/action/commands.py) - Execute a block of Python code interactively (in Jupyter notebook) and receives `CmdOutputObservation`. Requires setting up `jupyter` [plugin](../opendevin/runtime/plugins) as a requirement.
- [`FileReadAction`](../opendevin/events/action/files.py) - Reads the content of a file
- [`FileWriteAction`](../opendevin/events/action/files.py) - Writes new content to a file
- [`BrowseURLAction`](../opendevin/events/action/browse.py) - Gets the content of a URL
- [`AddTaskAction`](../opendevin/events/action/tasks.py) - Adds a subtask to the plan
- [`ModifyTaskAction`](../opendevin/events/action/tasks.py) - Changes the state of a subtask.
- [`AgentFinishAction`](../opendevin/events/action/agent.py) - Stops the control loop, allowing the user/delegator agent to enter a new task
- [`AgentRejectAction`](../opendevin/events/action/agent.py) - Stops the control loop, allowing the user/delegator agent to enter a new task
- [`AgentFinishAction`](../opendevin/events/action/agent.py) - Stops the control loop, allowing the user to enter a new task
- [`MessageAction`](../opendevin/events/action/message.py) - Represents a message from an agent or the user
You can use `action.to_dict()` and `action_from_dict` to serialize and deserialize actions.
## Observations
There are also several types of Observations. These are typically available in the step following the corresponding Action.
But they may also appear as a result of asynchronous events (e.g. a message from the user).
This folder implements the CodeAct idea ([paper](https://arxiv.org/abs/2402.01030), [tweet](https://twitter.com/xingyaow_/status/1754556835703751087)) that consolidates LLM agents’**act**ions into a unified **code** action space for both *simplicity* and *performance* (see paper for more details).
The conceptual idea is illustrated below. At each turn, the agent can:
1.**Converse**: Communicate with humans in natural language to ask for clarification, confirmation, etc.
2.**CodeAct**: Choose to perform the task by executing code
- Execute any valid Linux `bash` command
- Execute any valid `Python` code with [an interactive Python interpreter](https://ipython.org/). This is simulated through `bash` command, see plugin system below for more details.
To make the CodeAct agent more powerful with only access to `bash` action space, CodeAct agent leverages OpenDevin's plugin system:
- [Jupyter plugin](https://github.com/OpenDevin/OpenDevin/tree/main/opendevin/runtime/plugins/jupyter): for IPython execution via bash command
- [SWE-agent tool plugin](https://github.com/OpenDevin/OpenDevin/tree/main/opendevin/runtime/plugins/swe_agent_commands): Powerful bash command line tools for software development tasks introduced by [swe-agent](https://github.com/princeton-nlp/swe-agent).
The agent works by passing the model a list of action-observation pairs and prompting the model to take the next step.
### Overview
This agent implements the CodeAct idea ([paper](https://arxiv.org/abs/2402.13463), [tweet](https://twitter.com/xingyaow_/status/1754556835703751087)) that consolidates LLM agents’ **act**ions into a unified **code** action space for both *simplicity* and *performance* (see paper for more details).
The conceptual idea is illustrated below. At each turn, the agent can:
1. **Converse**: Communicate with humans in natural language to ask for clarification, confirmation, etc.
2. **CodeAct**: Choose to perform the task by executing code
- Execute any valid Linux `bash` command
- Execute any valid `Python` code with [an interactive Python interpreter](https://ipython.org/). This is simulated through `bash` command, see plugin system below for more details.
To make the CodeAct agent more powerful with only access to `bash` action space, CodeAct agent leverages OpenDevin's plugin system:
- [Jupyter plugin](https://github.com/OpenDevin/OpenDevin/tree/main/opendevin/runtime/plugins/jupyter): for IPython execution via bash command
- [SWE-agent tool plugin](https://github.com/OpenDevin/OpenDevin/tree/main/opendevin/runtime/plugins/swe_agent_commands): Powerful bash command line tools for software development tasks introduced by [swe-agent](https://github.com/princeton-nlp/swe-agent).
in_context_example:str=f"Here is an example of how you can interact with the environment for task solving:\n{get_in_context_example()}\n\nNOW, LET'S START!"
action_parser=CodeActResponseParser()
def__init__(
self,
llm:LLM,
)->None:
"""Initializes a new instance of the CodeActAgent class.
f'\n\nENVIRONMENT REMINDER: You have {state.max_iterations-state.iteration} turns left to complete the task. When finished reply with <finish></finish>'
'\nApart from the standard Python library, the assistant can also use the following functions (already imported) in <execute_ipython> environment:\n'
f'{_AGENT_SKILLS_DOCS}'
"Please note that THE `edit_file_by_replace`, `append_file` and `insert_content_at_line` FUNCTIONS REQUIRE PROPER INDENTATION. If the assistant would like to add the line ' print(x)', it must fully write that out, with all those spaces before the code! Indentation is important and code that is not indented correctly will fail and require fixing before it can be run."
)
# ======= SYSTEM MESSAGE =======
MINIMAL_SYSTEM_PREFIX="""A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.
The assistant can use an interactive Python (Jupyter Notebook) environment, executing code with <execute_ipython>.
<execute_ipython>
print("Hello World!")
</execute_ipython>
The assistant can execute bash commands on behalf of the user by wrapping them with <execute_bash> and </execute_bash>.
For example, you can list the files in the current directory by <execute_bash> ls </execute_bash>.
Important, however: do not run interactive commands. You do not have access to stdin.
Also, you need to handle commands that may run indefinitely and not return a result. For such cases, you should redirect the output to a file and run the command in the background to avoid blocking the execution.
For example, to run a Python script that might run indefinitely without returning immediately, you can use the following format: <execute_bash> python3 app.py > server.log 2>&1 & </execute_bash>
Also, if a command execution result saying like: Command: "npm start" timed out. Sending SIGINT to the process, you should also retry with running the command in the background.
"""
BROWSING_PREFIX="""The assistant can browse the Internet with <execute_browse> and </execute_browse>.
For example, <execute_browse> Tell me the usa's president using google search </execute_browse>.
Or <execute_browse> Tell me what is in http://example.com </execute_browse>.
"""
PIP_INSTALL_PREFIX="""The assistant can install Python packages using the %pip magic command in an IPython environment by using the following syntax: <execute_ipython> %pip install [package needed] </execute_ipython> and should always import packages and define variables before starting to use them."""
If $GITHUB_TOKEN is not set, ask the user to set it."""
SYSTEM_SUFFIX="""Responses should be concise.
The assistant should attempt fewer things at a time instead of putting too many commands OR too much code in one "execute" block.
Include ONLY ONE <execute_ipython>, <execute_bash>, or <execute_browse> per response, unless the assistant is finished with the task or needs more input or action from the user in order to proceed.
If the assistant is finished with the task you MUST include <finish></finish> in your response.
IMPORTANT: Execute code using <execute_ipython>, <execute_bash>, or <execute_browse> whenever possible.
When handling files, try to use full paths and pwd to avoid errors.
"""
# ======= EXAMPLE MESSAGE =======
EXAMPLES="""
--- START OF EXAMPLE ---
USER: Create a list of numbers from 1 to 10, and display them in a web page at port 5000.
ASSISTANT:
Sure! Let me create a Python file `app.py`:
<execute_ipython>
create_file('app.py')
</execute_ipython>
USER:
OBSERVATION:
[File: /workspace/app.py (1 lines total)]
(this is the beginning of the file)
1|
(this is the end of the file)
[File app.py created.]
ASSISTANT:
Now I will write the Python code for starting a web server and save it to the file `app.py`:
<execute_ipython>
EDITED_CODE=\"\"\"from flask import Flask
app = Flask(__name__)
@app.route('/')
def index():
numbers = list(range(1, 11))
return str(numbers)
if __name__ == '__main__':
app.run(port=5000)\"\"\"
insert_content_at_line(
'app.py',
1,
EDITED_CODE,
)
</execute_ipython>
USER:
OBSERVATION:
(this is the beginning of the file)
1|from flask import Flask
2|app = Flask(__name__)
3|
4|@app.route('/')
5|def index():
6| numbers = list(range(1, 11))
7| return str(numbers)
8|
9|if __name__ == '__main__':
10| app.run(port=5000)
(this is the end of the file)
[File updated (edited at line 1). Please review the changes and make sure they are correct (correct indentation, no duplicate lines, etc). Edit the file again if necessary.]
ASSISTANT:
I have created a Python file `app.py` that will display a list of numbers from 1 to 10 when you run it. Let me run the Python file for you:
Now that Flask is installed, let me run the Python file again:
<execute_bash>
python3 app.py > server.log 2>&1 &
</execute_bash>
USER:
OBSERVATION:
[1] 124
ASSISTANT:
Let me check the server log again:
<execute_bash>
cat server.log
</execute_bash>
USER:
OBSERVATION:
* Serving Flask app 'app'
* Debug mode: off
WARNING: This is a development server. Do not use it in a production deployment. Use a production WSGI server instead.
* Running on http://127.0.0.1:5000
Press CTRL+C to quit
ASSISTANT:
The server is running on port 5000 with PID 124. You can access the list of numbers by visiting http://127.0.0.1:5000. If you have any further questions, feel free to ask!
USER: Now browse the newly started server's homepage and show me the content.
ASSISTANT:
Sure! Let me browse the server's homepage at http://127.0.0.1:5000:
<execute_browse>
Get the content on "http://127.0.0.1:5000"
</execute_browse>
USER:
OBSERVATION:
[1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
ASSISTANT:
The content of the server's homepage is "[1, 2, 3, 4, 5, 6, 7, 8, 9, 10]". If you have any further questions, feel free to ask!
USER: Now kill the server, make it display the numbers in a table format.
ASSISTANT:
Sure! Let me stop the server first:
<execute_bash>
kill 124
</execute_bash>
USER:
OBSERVATION:
[1]+ Terminated python3 app.py > server.log 2>&1
ASSISTANT:
The server has been stopped. Let me open the Python file and modify it.
<execute_ipython>
open_file('app.py')
</execute_ipython>
USER:
[File: /workspace/app.py (10 lines total)]
(this is the beginning of the file)
1|from flask import Flask
2|app = Flask(__name__)
3|
4|@app.route('/')
5|def index():
6| numbers = list(range(1, 11))
7| return str(numbers)
8|
9|if __name__ == '__main__':
10| app.run(port=5000)
(this is the end of the file)
ASSISTANT:
I should edit the file to display the numbers in a table format. I should include correct indentation. Let me update the file:
<execute_ipython>
edit_file_by_replace(
'app.py',
to_replace=" return str(numbers)",
new_content=" return '<table>' + ''.join([f'<tr><td>{i}</td></tr>' for i in numbers]) + '</table>'",
)
</execute_ipython>
USER:
Observation:
[File: /workspace/app.py (10 lines total after edit)]
(this is the beginning of the file)
1|from flask import Flask
2|app = Flask(__name__)
3|
4|@app.route('/')
5|def index():
6| numbers = list(range(1, 11))
7| return '<table>' + ''.join([f'<tr><td>{i}</td></tr>' for i in numbers]) + '</table>'
8|
9|if __name__ == '__main__':
10| app.run(port=5000)
(this is the end of the file)
[File updated (edited at line 7). Please review the changes and make sure they are correct (correct indentation, no duplicate lines, etc). Edit the file again if necessary.]
ASSISTANT:
Running the updated file:
<execute_bash>
python3 app.py > server.log 2>&1 &
</execute_bash>
USER:
Observation:
[1] 126
ASSISTANT:
The server is running on port 5000 with PID 126. You can access the list of numbers in a table format by visiting http://127.0.0.1:5000. Let me know if you have any further requests!
<finish></finish>
--- END OF EXAMPLE ---
"""
INVALID_INPUT_MESSAGE=(
"I don't understand your input. \n"
'For bash commands, use <execute_bash> YOUR_COMMAND </execute_bash>.\n'
'For Python code, use <execute_ipython> YOUR_CODE </execute_ipython>.\n'
'For browsing, use <execute_browse> YOUR_COMMAND </execute_browse>.\n'
This agent is an adaptation of the original [SWE Agent](https://swe-agent.com/) based on CodeAct 1.5 using the `agentskills` library of OpenDevin.
It is intended use is **solving Github issues**.
It removes web-browsing and Github capability from the original CodeAct agent to avoid confusion to the agent.
"""
sandbox_plugins:list[PluginRequirement]=[
# NOTE: AgentSkillsRequirement need to go before JupyterRequirement, since
# AgentSkillsRequirement provides a lot of Python functions,
# and it needs to be initialized before Jupyter for Jupyter to use those functions.
AgentSkillsRequirement(),
JupyterRequirement(),
]
runtime_tools:list[RuntimeTool]=[]
system_message:str=get_system_message()
in_context_example:str=f"Here is an example of how you can interact with the environment for task solving:\n{get_in_context_example()}\n\nNOW, LET'S START!"
response_parser=CodeActSWEResponseParser()
def__init__(
self,
llm:LLM,
)->None:
"""Initializes a new instance of the CodeActAgent class.
"""Modifies the current state by adding the most recent actions and observations, then prompts the model to think about it's next action to take using monologue, memory, and hint.
Parameters:
- state (State): The current state based on previous steps taken
Returns:
- Action: The next action to take based on LLM response
You're a thoughtful robot. Your main task is this:
%(task)s
Don't expand the scope of your task--just complete it as written.
This is your internal monologue, in JSON format:
%(monologue)s
Your most recent thought is at the bottom of that monologue. Continue your train of thought.
What is your next single thought or action? Your response must be in JSON format.
It must be a single object, and it must contain two fields:
* `action`, which is one of the actions below
* `args`, which is a map of key-value pairs, specifying the arguments for that action
Here are the possible actions:
* `read` - reads the content of a file. Arguments:
* `path` - the path of the file to read
* `write` - writes the content to a file. Arguments:
* `path` - the path of the file to write
* `content` - the content to write to the file
* `run` - runs a command. Arguments:
* `command` - the command to run
* `browse` - opens a web page. Arguments:
* `url` - the URL to open
* `push` - Push a branch from the current repo to github:
* `owner` - the owner of the repo to push to
* `repo` - the name of the repo to push to
* `branch` - the name of the branch to push
* `message` - make a plan, set a goal, record your thoughts, or ask for more input from the user. Arguments:
* `content` - the message to record
* `wait_for_response` - set to `true` to wait for the user to respond before proceeding
* `finish` - if you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, and push actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.
Notes:
* you are logged in as %(user)s, but sudo will always work without a password.
* all non-background commands will be forcibly stopped if they remain running for over %(timeout)s seconds.
* your environment is Debian Linux. You can install software with `sudo apt-get`, but remember to use -y.
* don't run interactive commands, or commands that don't return (e.g. `node server.js`). You may run commands in the background (e.g. `node server.js &`)
* don't run interactive text editors (e.g. `nano` or 'vim'), instead use the 'write' or 'read' action.
* don't run gui applications (e.g. software IDEs (like vs code or codium), web browsers (like firefox or chromium), or other complex software packages). Use non-interactive cli applications, or special actions instead.
* whenever an action fails, always send a `message` about why it may have happened before acting again.
What is your next single thought or action? Again, you must reply with JSON, and only with JSON. You must respond with exactly one 'action' object.
%(hint)s
"""
MONOLOGUE_SUMMARY_PROMPT="""
Below is the internal monologue of an automated LLM agent. Each
thought is an item in a JSON array. The thoughts may be memories,
actions taken by the agent, or outputs from those actions.
Please return a new, smaller JSON array, which summarizes the
internal monologue. You can summarize individual thoughts, and
you can condense related thoughts together with a description
of their content.
%(monologue)s
Make the summaries as pithy and informative as possible.
Be specific about what happened and what was learned. The summary
will be used as keywords for searching for the original memory.
Be sure to preserve any key words or important information.
Your response must be in JSON format. It must be an object with the
key `new_monologue`, which is a JSON array containing the summarized monologue.
Each entry in the array must have an `action` key, and an `args` key.
The action key may be `summarize`, and `args.summary` should contain the summary.
You can also use the same action and args from the source monologue.
"""
INITIAL_THOUGHTS=[
'I exist!',
'Hmm...looks like I can type in a command line prompt',
'Looks like I have a web browser too!',
"Here's what I want to do: $TASK",
'How am I going to get there though?',
'It seems like I have some kind of short term memory.',
'Each of my thoughts seems to be stored in a JSON array.',
'It seems whatever I say next will be added as an object to the list.',
"It looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
'RUN echo "hello world"',
'hello world',
'Cool! I bet I can write files too using the write action.',
Install [Docker](https://docs.docker.com/engine/install/) on your host machine and run:
```bash
make docker-dev
# same as:
cd ./containers/dev
./dev.sh
```
It could take some time if you are running for the first time as Docker will pull all the tools required for building OpenHands. The next time you run again, it should be instant.
## Build and run
If everything goes well, you should be inside a container after Docker finishes building the `openhands:dev` image similar to the following:
```bash
Build and run in Docker ...
root@93fc0005fcd2:/app#
```
You may now proceed with the normal [build and run](../../Development.md) workflow as if you were on the host.
## Make changes
The source code on the host is mounted as `/app` inside docker. You may edit the files as usual either inside the Docker container or on your host with your favorite IDE/editors.
The following are also mapped as readonly from your host:
[E2B](https://e2b.dev) is an [open-source](https://github.com/e2b-dev/e2b) secure cloud environment (sandbox) made for running AI-generated code and agents. E2B offers [Python](https://pypi.org/project/e2b/) and [JS/TS](https://www.npmjs.com/package/e2b) SDK to spawn and control these sandboxes.
This folder builds a runtime image (sandbox), which will use a dynamically generated `Dockerfile`
that depends on the `base_image`**AND** a [Python source distribution](https://docs.python.org/3.10/distutils/sourcedist.html) that is based on the current commit of `openhands`.
The following command will generate a `Dockerfile` file for `nikolaik/python-nodejs:python3.12-nodejs22` (the default base image), an updated `config.sh` and the runtime source distribution files/folders into `containers/runtime`:
```bash
poetry run python3 openhands/runtime/utils/runtime_build.py \
"message":"est un ingénieur logiciel autonome qui peut résoudre des tâches d'ingénierie logicielle et de navigation web à tout moment. Il peut exécuter des requêtes en sciences des données, telles que \"Trouver le nombre de demandes de pull à l'repository OpenDevin dans les derniers mois\", et des tâches d'ingénierie logicielle, comme \"Veuillez ajouter des tests à ce fichier et vérifier si tous les tests passent. Si ce n'est pas le cas, réparez le fichier.\"",
"description":"Description for OpenDevin"
"message":"est un ingénieur logiciel autonome qui peut résoudre des tâches d'ingénierie logicielle et de navigation web à tout moment. Il peut exécuter des requêtes en sciences des données, telles que \"Trouver le nombre de demandes de pull à l'repository OpenHands dans les derniers mois\", et des tâches d'ingénierie logicielle, comme \"Veuillez ajouter des tests à ce fichier et vérifier si tous les tests passent. Si ce n'est pas le cas, réparez le fichier.\"",
"description":"Description for OpenHands"
},
"faq.section.description.2":{
"message":"De plus, OpenDevin est une plateforme et communauté pour les développeurs d'agents qui souhaitent tester et évaluer de nouveaux agents.",
"description":"Further Description for OpenDevin"
"message":"De plus, OpenHands est une plateforme et communauté pour les développeurs d'agents qui souhaitent tester et évaluer de nouveaux agents.",
"description":"Further Description for OpenHands"
},
"faq.section.title.2":{
"message":"Support",
@@ -43,15 +43,15 @@
"description":"Support Answer"
},
"faq.section.title.3":{
"message":"Comment résoudre un problème sur GitHub avec OpenDevin ?",
"message":"Comment résoudre un problème sur GitHub avec OpenHands ?",
"description":"GitHub Issue Section Title"
},
"faq.section.github.steps.intro":{
"message":"Pour résoudre un problème sur GitHub en utilisant OpenDevin, envoyez une commande à OpenDevin demandant qu'il suit des étapes comme les suivantes :",
"message":"Pour résoudre un problème sur GitHub en utilisant OpenHands, envoyez une commande à OpenHands demandant qu'il suit des étapes comme les suivantes :",
"message":"Avant de lancer OpenDevin, vous pouvez faire :",
"message":"Avant de lancer OpenHands, vous pouvez faire :",
"description":"GitHub Steps Pre-Run"
},
"faq.section.github.steps.tokenInfo":{
"message":"où XXX est un jeton GitHub que vous avez créé et qui a les autorisations pour pousser dans le dépôt OpenDevin. Si vous n'avez pas d'autorisations de modification du dépôt OpenDevin, vous devrez peut-être changer cela en :",
"message":"où XXX est un jeton GitHub que vous avez créé et qui a les autorisations pour pousser dans le dépôt OpenHands. Si vous n'avez pas d'autorisations de modification du dépôt OpenHands, vous devrez peut-être changer cela en :",
"description":"GitHub Steps Token Info"
},
"faq.section.github.steps.usernameInfo":{
@@ -83,23 +83,23 @@
"description":"GitHub Steps Username Info"
},
"faq.section.title.4":{
"message":"Comment OpenDevin est-il différent de Devin ?",
"message":"Comment OpenHands est-il différent de Devin ?",
"description":"Devin Section Title"
},
"faq.section.devin.linkText":{
"faq.section.openhands.linkText":{
"message":"Devin",
"description":"Devin Link Text"
},
"faq.section.devin.description":{
"message":"est un produit commercial par Cognition Inc., qui a servi d'inspiration initiale pour OpenDevin. Les deux visent à bien faire le travail d'ingénierie logicielle, mais vous pouvez télécharger, utiliser et modifier OpenDevin, tandis que Devin peut être utilisé uniquement via le site de Cognition. De plus, OpenDevin a évolué au-delà de l'inspiration initiale, et est maintenant un écosystème communautaire pour le développement d'agents en général, et nous serions ravis de vous voir rejoindre et",
"faq.section.openhands.description":{
"message":"est un produit commercial par Cognition Inc., qui a servi d'inspiration initiale pour OpenHands. Les deux visent à bien faire le travail d'ingénierie logicielle, mais vous pouvez télécharger, utiliser et modifier OpenHands, tandis que Devin peut être utilisé uniquement via le site de Cognition. De plus, OpenHands a évolué au-delà de l'inspiration initiale, et est maintenant un écosystème communautaire pour le développement d'agents en général, et nous serions ravis de vous voir rejoindre et",
"description":"Devin Description"
},
"faq.section.devin.contribute":{
"faq.section.openhands.contribute":{
"message":"contribuer",
"description":"Contribute Link"
},
"faq.section.title.5":{
"message":"Comment OpenDevin est-il différent de ChatGPT ?",
"message":"Comment OpenHands est-il différent de ChatGPT ?",
"description":"ChatGPT Section Title"
},
"faq.section.chatgpt.description":{
@@ -114,7 +114,7 @@
"message":"Commencer"
},
"welcome.message":{
"message":"Bienvenue à OpenDevin, un système d'IA autonome ingénieur logiciel capable d'exécuter des tâches d'ingénierie complexes et de collaborer activement avec les utilisateurs sur les projets de développement logiciel."
"message":"Bienvenue à OpenHands, un système d'IA autonome ingénieur logiciel capable d'exécuter des tâches d'ingénierie complexes et de collaborer activement avec les utilisateurs sur les projets de développement logiciel."
La réplication complète d'applications de niveau production avec des LLM est une entreprise complexe. Notre stratégie implique :
La réalisation d'une réplication complète des applications de production avec les LLM est une entreprise complexe. Notre stratégie implique :
1.**Recherche technique fondamentale :** Se concentrer sur la recherche fondamentale pour comprendre et améliorer les aspects techniques de la génération et de la gestion de code
2.**Capacités spécialisées :** Améliorer l'efficacité des composants de base grâce à la curation de données, aux méthodes d'entraînement, etc.
3.**Planification des tâches :** Développer des capacités de détection de bugs, de gestion de base de code et d'optimisation
4.**Évaluation :** Établir des métriques d'évaluation complètes pour mieux comprendre et améliorer nos modèles
1.**Recherche Technique de Base :** Se concentrer sur la recherche fondamentale pour comprendre et améliorer les aspects techniques de la génération et de la gestion de code.
2.**Compétences Spécialisées :** Améliorer l'efficacité des composants de base grâce à la curation des données, aux méthodes de formation, et plus encore.
3.**Planification des Tâches :** Développer des capacités pour la détection de bogues, la gestion du code source et l'optimisation.
4.**Évaluation :** Établir des métriques d'évaluation complètes pour mieux comprendre et améliorer nos modèles.
## 🚧 Agent par défaut
## 🚧 Agent Par Défaut {#default-agent}
Notre Agent par défaut est actuellement le [CodeActAgent](agents), qui est capable de générer du code et de gérer des fichiers.
- Notre agent par défaut est actuellement le CodeActAgent, capable de générer du code et de gérer des fichiers. Nous travaillons sur d'autres implémentations d'agents, y compris [SWE Agent](https://swe-agent.com/). Vous pouvez [lire à propos de notre ensemble actuel d'agents ici](./agents).
## 🤝 Comment contribuer
## 🤝 Comment Contribuer {#how-to-contribute}
OpenHands est un projet communautaire et nous accueillons les contributions de tous. Que vous soyez développeur, chercheur ou simplement enthousiaste à l'idée de faire progresser le domaine de l'ingénierie logicielle avec l'IA, il existe de nombreuses façons de s'impliquer :
OpenDevin est un projet communautaire, et nous accueillons les contributions de tout le monde. Que vous soyez développeur, chercheur, ou simplement enthousiaste à l'idée de faire progresser le domaine de l'ingénierie logicielle avec l'IA, il existe de nombreuses façons de vous impliquer :
- **Contributions de code :** Aidez-nous à développer les fonctionnalités de base, l'interface frontend ou les solutions de sandboxing
- **Recherche et évaluation :** Contribuez à notre compréhension des LLM dans l'ingénierie logicielle, participez à l'évaluation des modèles ou suggérez des améliorations
- **Retours et tests :** Utilisez la boîte à outils OpenHands, signalez des bugs, suggérez des fonctionnalités ou donnez votre avis sur la facilité d'utilisation
- **Contributions de Code :** Aidez-nous à développer les fonctionnalités de base, l'interface frontend ou les solutions de sandboxing.
- **Recherche et Évaluation :** Contribuez à notre compréhension des LLM en ingénierie logicielle, participez à l'évaluation des modèles ou suggérez des améliorations.
- **Retour d'Information et Tests :** Utilisez l'ensemble d'outils OpenDevin, signalez des bogues, suggérez des fonctionnalités ou fournissez des retours sur l'ergonomie.
Pour plus de détails, veuillez consulter [ce document](https://github.com/All-Hands-AI/OpenHands/blob/main/CONTRIBUTING.md).
Pour plus de détails, veuillez consulter [ce document](https://github.com/OpenDevin/OpenDevin/blob/main/CONTRIBUTING.md).
Nous avons à la fois un espace de travail Slack pour la collaboration sur la construction d'OpenHands et un serveur Discord pour discuter de tout ce qui est lié, par exemple, à ce projet, LLM, agent, etc.
Nous avons maintenant à la fois un espace de travail Slack pour la collaboration sur la construction d'OpenDevin et un serveur Discord pour discuter de tout ce qui est lié, par exemple, à ce projet, aux LLM, aux agents, etc.
- [Espace de travail Slack](https://join.slack.com/t/opendevin/shared_invite/zt-2jsrl32uf-fTeeFjNyNYxqSZt5NPY3fA)
- [Espace de travail Slack](https://join.slack.com/t/opendevin/shared_invite/zt-2oikve2hu-UDxHeo8nsE69y6T7yFX_BA)
Si vous souhaitez contribuer, n'hésitez pas à rejoindre notre communauté. Simplifions l'ingénierie logicielle ensemble !
Si vous souhaitez contribuer, n'hésitez pas à rejoindre notre communauté. Simplifions ensemble l'ingénierie logicielle !
🐚 **Codez moins, créez plus avec OpenDevin.**
🐚 **Codez moins, faites plus avec OpenHands.**
[](https://star-history.com/#OpenDevin/OpenDevin&Date)
[](https://star-history.com/#All-Hands-AI/OpenHands&Date)
## 🛠️ Construit Avec {#built-with}
## 🛠️ Construit avec
OpenDevin est construit en utilisant une combinaison de cadres et de bibliothèques puissants, offrant une base robuste pour son développement. Voici les technologies clés utilisées dans le projet :
OpenHands est construit en utilisant une combinaison de frameworks et de bibliothèques puissants, fournissant une base solide pour son développement. Voici les principales technologies utilisées dans le projet :
Veuillez noter que la sélection de ces technologies est en cours, et que des technologies supplémentaires peuvent être ajoutées ou des existantes supprimées au fur et à mesure de l'évolution du projet. Nous nous efforçons d'adopter les outils les plus adaptés et efficaces pour améliorer les capacités d'OpenDevin.
Veuillez noter que la sélection de ces technologies est en cours et que des technologies supplémentaires peuvent être ajoutées ou des technologies existantes peuvent être supprimées à mesure que le projet évolue. Nous nous efforçons d'adopter les outils les plus appropriés et les plus efficaces pour améliorer les capacités d'OpenHands.
## 📜 Licence {#license}
## 📜 Licence
Distribué sous la licence MIT. Voir [notre licence](https://github.com/OpenDevin/OpenDevin/blob/main/LICENSE) pour plus d'informations.
Distribué sous la licence MIT. Voir [notre licence](https://github.com/All-Hands-AI/OpenHands/blob/main/LICENSE) pour plus d'informations.
Cet agent implémente l'idée CodeAct ([article](https://arxiv.org/abs/2402.01030), [tweet](https://twitter.com/xingyaow_/status/1754556835703751087)) qui consolide les **act**ions des agents LLM en un espace d'action **code** unifié pour à la fois la _simplicité_ et la _performance_ (voir article pour plus de détails).
Cet agent implémente l'idée de CodeAct ([article](https://arxiv.org/abs/2402.01030), [tweet](https://twitter.com/xingyaow_/status/1754556835703751087)) qui consolide les **act**ions des agents LLM dans un espace d'action de **code** unifié à la fois pour la _simplicité_ et la _performance_.
L'idée conceptuelle est illustrée ci-dessous. À chaque tour, l'agent peut :
1.**Converse** : Communiquer avec les humains en langage naturel pour demander des clarifications, des confirmations, etc.
2.**CodeAct** : Choisir d'accomplir la tâche en exécutant du code
1.**Converser** : Communiquer avec les humains en langage naturel pour demander des clarifications, des confirmations, etc.
2.**CodeAct** : Choisir d'effectuer la tâche en exécutant du code
- Exécuter toute commande `bash` Linux valide
- Exécuter tout code `Python` valide avec [un interpréteur Python interactif](https://ipython.org/). Cela est simulé à travers la commande `bash`, voir le système de plugin ci-dessous pour plus de détails.
- Exécuter n'importe quelle commande Linux `bash` valide
- Exécuter n'importe quel code `Python` valide avec [un interpréteur Python interactif](https://ipython.org/). Ceci est simulé via une commande `bash`, voir le système de plugin ci-dessous pour plus de détails.
Pour rendre l'agent CodeAct plus puissant avec seulement l'accès à l'espace d'action `bash`, l'agent CodeAct exploite le système de plugins d'OpenDevin:
- [Plugin Jupyter](https://github.com/OpenDevin/OpenDevin/tree/main/opendevin/runtime/plugins/jupyter) : pour l'exécution d'IPython via la commande bash
- [Plugin outil agent SWE](https://github.com/OpenDevin/OpenDevin/tree/main/opendevin/runtime/plugins/swe_agent_commands) : Outils de ligne de commande bash puissants pour les tâches de développement logiciel introduits par [swe-agent](https://github.com/princeton-nlp/swe-agent).
| `__init__` | Initialise un agent avec `llm` et une liste de messages `list[Mapping[str, str]]` |
| `step` | Effectue une étape en utilisant l'agent CodeAct. Cela inclut la collecte d'informations sur les étapes précédentes et invite le modèle à exécuter une commande. |
### En cours de réalisation & prochaine étape
[] Support de la navigation sur le web
[] Compléter le workflow pour l'agent CodeAct afin de soumettre des PRs Github
## Agent Monologue
### Description
L'agent Monologue utilise la mémoire à long terme et à court terme pour accomplir des tâches.
La mémoire à long terme est stockée en tant qu'objet LongTermMemory et le modèle l'utilise pour rechercher des exemples du passé.
La mémoire à court terme est stockée en tant qu'objet Monologue et le modèle peut la condenser si nécessaire.
| `__init__` | Initialise l'agent avec une mémoire à long terme et un monologue interne |
| `_add_event` | Ajoute des événements au monologue de l'agent et condense avec un résumé automatiquement si le monologue est trop long |
| `_initialize` | Utilise la liste `INITIAL_THOUGHTS` pour donner à l'agent un contexte pour ses capacités et comment naviguer dans le `/workspace` |
| `step` | Modifie l'état actuel en ajoutant les actions et observations les plus récentes, puis invite le modèle à réfléchir à la prochaine action à entreprendre. |
## Agent Planificateur
### Description
L'agent planificateur utilise une stratégie d'incitation spéciale pour créer des plans à long terme pour résoudre les problèmes.
L'agent reçoit ses paires action-observation précédentes, la tâche actuelle, et un indice basé sur la dernière action effectuée à chaque étape.
| `step` | Vérifie si l'étape actuelle est terminée, retourne `AgentFinishAction` si oui. Sinon, crée une incitation de planification et l'envoie au modèle pour inférence, en ajoutant le résultat comme prochaine action. |
_Exemple de CodeActAgent avec `gpt-4-turbo-2024-04-09` effectuant une tâche de science des données (régression linéaire)_.
2. Ouvrez le fichier généré dans un éditeur PlantUML, par exemple Visual Studio Code avec l'extension PlantUML ou [PlantText](https://www.planttext.com/)
<img src="https://github.com/All-Hands-AI/OpenHands/assets/16201837/97d747e3-29d8-4ccb-8d34-6ad1adb17f38" alt="OpenHands System Architecture Diagram Jul 4 2024" />
<p><em>Diagramme de l'Architecture du Système OpenHands (4 juillet 2024)</em></p>
</div>
Ceci est une vue d'ensemble de haut niveau de l'architecture du système. Le système est divisé en deux composants principaux : le frontend et le backend. Le frontend est responsable de la gestion des interactions utilisateur et de l'affichage des résultats. Le backend est responsable de la gestion de la logique métier et de l'exécution des agents.
# Architecture du Frontend {#frontend-architecture-fr}
Cette vue d'ensemble est simplifiée pour montrer les principaux composants et leurs interactions. Pour une vue plus détaillée de l'architecture du backend, voir la section Architecture du Backend ci-dessous.
# Architecture du Backend {#backend-architecture-fr}
_**Avertissement** : L'architecture du backend est en cours de développement et est sujette à changement. Le diagramme suivant montre l'architecture actuelle du backend basée sur le commit indiqué dans le pied de page du diagramme._
La génération du diagramme d'architecture du backend est partiellement automatisée.
Le diagramme est généré à partir des indications de type dans le code en utilisant l'outil py2puml. Le diagramme est ensuite manuellement revu, ajusté et exporté en PNG et SVG.
## Prérequis
- Environnement python fonctionnel dans lequel openhands est exécutable
(selon les instructions du fichier README.md à la racine du dépôt)
2. Ouvrir le fichier généré dans un éditeur PlantUML, par ex. Visual Studio Code avec l'extension PlantUML ou [PlantText](https://www.planttext.com/)
3. Revoir le PUML généré et effectuer tous les ajustements nécessaires au diagramme (ajouter les parties manquantes, corriger les erreurs, améliorer le positionnement).
_py2puml crée le diagramme en se basant sur les indications de type dans le code, donc des indications manquantes ou incorrectes peuvent entraîner un diagramme incomplet ou incorrect._
4. Revoir la différence entre le nouveau diagramme et le précédent et vérifier manuellement si les changements sont corrects.
_S'assurer de ne pas supprimer des parties qui ont été ajoutées manuellement au diagramme par le passé et qui sont toujours pertinentes._
5. Ajouter le hash du commit qui a été utilisé pour générer le diagramme dans le pied de page du diagramme.
6. Exporter le diagramme sous forme de fichiers PNG et SVG et remplacer les diagrammes existants dans le répertoire `docs/architecture`. Cela peut être fait avec (par ex. [PlantText](https://www.planttext.com/))
Le Runtime EventStream d'OpenHands est le composant principal qui permet l'exécution sécurisée et flexible des actions des agents d'IA.
Il crée un environnement en bac à sable (sandbox) en utilisant Docker, où du code arbitraire peut être exécuté en toute sécurité sans risquer le système hôte.
## Pourquoi avons-nous besoin d'un runtime en bac à sable ?
OpenHands doit exécuter du code arbitraire dans un environnement sécurisé et isolé pour plusieurs raisons :
1. Sécurité : L'exécution de code non fiable peut poser des risques importants pour le système hôte. Un environnement en bac à sable empêche le code malveillant d'accéder ou de modifier les ressources du système hôte
2. Cohérence : Un environnement en bac à sable garantit que l'exécution du code est cohérente sur différentes machines et configurations, éliminant les problèmes du type "ça fonctionne sur ma machine"
3. Contrôle des ressources : Le bac à sable permet un meilleur contrôle de l'allocation et de l'utilisation des ressources, empêchant les processus incontrôlés d'affecter le système hôte
4. Isolation : Différents projets ou utilisateurs peuvent travailler dans des environnements isolés sans interférer les uns avec les autres ou avec le système hôte
5. Reproductibilité : Les environnements en bac à sable facilitent la reproduction des bugs et des problèmes, car l'environnement d'exécution est cohérent et contrôlable
## Comment fonctionne le Runtime ?
Le système Runtime d'OpenHands utilise une architecture client-serveur implémentée avec des conteneurs Docker. Voici un aperçu de son fonctionnement :
```mermaid
graph TD
A[Image Docker personnalisée fournie par l'utilisateur] --> B[Backend OpenHands]
B -->|Construit| C[Image OH Runtime]
C -->|Lance| D[Exécuteur d'actions]
D -->|Initialise| E[Navigateur]
D -->|Initialise| F[Shell Bash]
D -->|Initialise| G[Plugins]
G -->|Initialise| L[Serveur Jupyter]
B -->|Génère| H[Agent]
B -->|Génère| I[EventStream]
I <--->|Exécute l'action pour
obtenir l'observation
via l'API REST
| D
H -->|Génère l'action| I
I -->|Obtient l'observation| H
subgraph "Conteneur Docker"
D
E
F
G
L
end
```
1. Entrée utilisateur : L'utilisateur fournit une image Docker de base personnalisée
2. Construction de l'image : OpenHands construit une nouvelle image Docker (l'"image OH runtime") basée sur l'image fournie par l'utilisateur. Cette nouvelle image inclut le code spécifique à OpenHands, principalement le "client runtime"
3. Lancement du conteneur : Lorsqu'OpenHands démarre, il lance un conteneur Docker en utilisant l'image OH runtime
4. Initialisation du serveur d'exécution des actions : Le serveur d'exécution des actions initialise un `ActionExecutor` à l'intérieur du conteneur, mettant en place les composants nécessaires comme un shell bash et chargeant les plugins spécifiés
5. Communication : Le backend OpenHands (`openhands/runtime/impl/eventstream/eventstream_runtime.py`) communique avec le serveur d'exécution des actions via une API RESTful, envoyant des actions et recevant des observations
6. Exécution des actions : Le client runtime reçoit les actions du backend, les exécute dans l'environnement en bac à sable et renvoie les observations
7. Retour des observations : Le serveur d'exécution des actions renvoie les résultats d'exécution au backend OpenHands sous forme d'observations
Le rôle du client :
- Il agit comme un intermédiaire entre le backend OpenHands et l'environnement en bac à sable
- Il exécute différents types d'actions (commandes shell, opérations sur les fichiers, code Python, etc.) en toute sécurité dans le conteneur
- Il gère l'état de l'environnement en bac à sable, y compris le répertoire de travail courant et les plugins chargés
- Il formate et renvoie les observations au backend, assurant une interface cohérente pour le traitement des résultats
## Comment OpenHands construit et maintient les images OH Runtime
L'approche d'OpenHands pour la construction et la gestion des images runtime assure l'efficacité, la cohérence et la flexibilité dans la création et la maintenance des images Docker pour les environnements de production et de développement.
Consultez le [code pertinent](https://github.com/All-Hands-AI/OpenHands/blob/main/openhands/runtime/utils/runtime_build.py) si vous souhaitez plus de détails.
### Système de balises d'images
OpenHands utilise un système à trois balises pour ses images runtime afin d'équilibrer la reproductibilité et la flexibilité.
Les balises peuvent être dans l'un des 2 formats suivants :
- **Balise versionnée** : `oh_v{openhands_version}_{base_image}` (ex : `oh_v0.9.9_nikolaik_s_python-nodejs_t_python3.12-nodejs22`)
- **Balise de verrouillage** : `oh_v{openhands_version}_{16_digit_lock_hash}` (ex : `oh_v0.9.9_1234567890abcdef`)
(ex : `oh_v0.9.9_1234567890abcdef_1234567890abcdef`)
#### Balise source - La plus spécifique
Il s'agit des 16 premiers chiffres du MD5 du hash du répertoire pour le répertoire source. Cela donne un hash
uniquement pour la source d'openhands
#### Balise de verrouillage
Ce hash est construit à partir des 16 premiers chiffres du MD5 de :
- Le nom de l'image de base sur laquelle l'image a été construite (ex : `nikolaik/python-nodejs:python3.12-nodejs22`)
- Le contenu du `pyproject.toml` inclus dans l'image.
- Le contenu du `poetry.lock` inclus dans l'image.
Cela donne effectivement un hash pour les dépendances d'Openhands indépendamment du code source.
#### Balise versionnée - La plus générique
Cette balise est une concaténation de la version d'openhands et du nom de l'image de base (transformé pour s'adapter au standard des balises).
#### Processus de construction
Lors de la génération d'une image...
- **Pas de reconstruction** : OpenHands vérifie d'abord si une image avec la même **balise source la plus spécifique** existe. S'il existe une telle image,
aucune construction n'est effectuée - l'image existante est utilisée.
- **Reconstruction la plus rapide** : OpenHands vérifie ensuite si une image avec la **balise de verrouillage générique** existe. S'il existe une telle image,
OpenHands construit une nouvelle image basée sur celle-ci, en contournant toutes les étapes d'installation (comme `poetry install` et
`apt-get`) sauf une opération finale pour copier le code source actuel. La nouvelle image est balisée avec une
balise **source** uniquement.
- **Reconstruction correcte** : Si ni une balise **source** ni une balise **de verrouillage** n'existe, une image sera construite sur la base de l'image avec la balise **versionnée**.
Dans l'image avec la balise versionnée, la plupart des dépendances devraient déjà être installées, ce qui permet de gagner du temps.
- **Reconstruction la plus lente** : Si les trois balises n'existent pas, une toute nouvelle image est construite à partir de
l'image de base (ce qui est une opération plus lente). Cette nouvelle image est balisée avec toutes les balises **source**, **de verrouillage** et **versionnée**.
Cette approche de balisage permet à OpenHands de gérer efficacement les environnements de développement et de production.
1. Un code source et un Dockerfile identiques produisent toujours la même image (via des balises basées sur des hashs)
2. Le système peut reconstruire rapidement les images lorsque des changements mineurs se produisent (en s'appuyant sur des images compatibles récentes)
3. La balise **de verrouillage** (ex : `runtime:oh_v0.9.3_1234567890abcdef`) pointe toujours vers la dernière version pour une combinaison particulière d'image de base, de dépendances et de version d'OpenHands
## Système de plugins du Runtime
Le Runtime d'OpenHands prend en charge un système de plugins qui permet d'étendre les fonctionnalités et de personnaliser l'environnement d'exécution. Les plugins sont initialisés lorsque le client runtime démarre.
Consultez [un exemple de plugin Jupyter ici](https://github.com/All-Hands-AI/OpenHands/blob/ecf4aed28b0cf7c18d4d8ff554883ba182fc6bdd/openhands/runtime/plugins/jupyter/__init__.py#L21-L55) si vous souhaitez implémenter votre propre plugin.
*Plus de détails sur le système de plugins sont encore en construction - les contributions sont les bienvenues !*
Aspects clés du système de plugins :
1. Définition des plugins : Les plugins sont définis comme des classes Python qui héritent d'une classe de base `Plugin`
2. Enregistrement des plugins : Les plugins disponibles sont enregistrés dans un dictionnaire `ALL_PLUGINS`
3. Spécification des plugins : Les plugins sont associés à `Agent.sandbox_plugins: list[PluginRequirement]`. Les utilisateurs peuvent spécifier quels plugins charger lors de l'initialisation du runtime
4. Initialisation : Les plugins sont initialisés de manière asynchrone lorsque le client runtime démarre
5. Utilisation : Le client runtime peut utiliser les plugins initialisés pour étendre ses capacités (par exemple, le JupyterPlugin pour exécuter des cellules IPython)
Le sandbox par défaut OpenDevin est équipé d'une configuration ubuntu minimaliste. Votre cas d'utilisation pourrait nécessiter des logiciels installés par défaut. Cet article vous enseignera comment réaliser cela en utilisant une image docker personnalisée.
Le sandbox par défaut OpenHands est équipé d'une configuration ubuntu minimaliste. Votre cas d'utilisation pourrait nécessiter des logiciels installés par défaut. Cet article vous enseignera comment réaliser cela en utilisant une image docker personnalisée.
## Configuration
Assurez-vous de pouvoir utiliser OpenDevin en suivant la documentation [Development.md](https://github.com/OpenDevin/OpenDevin/blob/main/Development.md).
Assurez-vous de pouvoir utiliser OpenHands en suivant la documentation [Development.md](https://github.com/All-Hands-AI/OpenHands/blob/main/Development.md).
## Créer Votre Image Docker
Ensuite, vous devez créer votre image docker personnalisée qui doit être basée sur debian/ubuntu. Par exemple, si nous souhaitons que OpenDevin ait accès au "node" binaire, nous utiliserions ce Dockerfile:
Ensuite, vous devez créer votre image docker personnalisée qui doit être basée sur debian/ubuntu. Par exemple, si nous souhaitons que OpenHands ait accès au "node" binaire, nous utiliserions ce Dockerfile:
Cela produira une nouvelle image appelée ```image_personnalisée``` qui sera disponible dans Docker Engine.
> Remarque: Dans la configuration décrite ici, OpenDevin va fonctionner en tant que utilisateur "opendevin" à l'intérieur du sandbox et donc tous les packages installés via le Dockerfile seront disponibles pour tous les utilisateurs sur le système, pas seulement root.
> Remarque: Dans la configuration décrite ici, OpenHands va fonctionner en tant que utilisateur "openhands" à l'intérieur du sandbox et donc tous les packages installés via le Dockerfile seront disponibles pour tous les utilisateurs sur le système, pas seulement root.
>
> L'installation avec apt-get ci-dessus installe nodejs pour tous les utilisateurs.
## Spécifiez votre image personnalisée dans le fichier config.toml
La configuration OpenDevin se fait via le fichier de niveau supérieur ```config.toml``` .
Créez un fichier ```config.toml``` dans le répertoire OpenDevin et entrez ces contenus:
La configuration OpenHands se fait via le fichier de niveau supérieur ```config.toml``` .
Créez un fichier ```config.toml``` dans le répertoire OpenHands et entrez ces contenus:
> Assurez-vous que ```sandbox_container_image``` est défini sur le nom de votre image personnalisée précédente.
> Assurez-vous que ```sandbox_base_container_image``` est défini sur le nom de votre image personnalisée précédente.
## Exécution
Exécutez OpenDevin en exécutant ```make run``` dans le répertoire racine.
Exécutez OpenHands en exécutant ```make run``` dans le répertoire racine.
Naviguez vers ```localhost:3001``` et vérifiez si vos dépendances souhaitées sont disponibles.
@@ -60,41 +59,36 @@ Félicitations !
## Explication technique
Le code pertinent est défini dans [ssh_box.py](https://github.com/OpenDevin/OpenDevin/blob/main/opendevin/runtime/docker/ssh_box.py) et [image_agnostic_util.py](https://github.com/OpenDevin/OpenDevin/blob/main/opendevin/runtime/docker/image_agnostic_util.py).
En particulier, ssh_box.py vérifie l'objet config pour ```config.sandbox_container_image``` et ensuite tente de récupérer l'image à l'aide de [get_od_sandbox_image](https://github.com/OpenDevin/OpenDevin/blob/main/opendevin/runtime/docker/image_agnostic_util.py#L72), qui est défini dans image_agnostic_util.py.
Lorsqu'une image personnalisée est utilisée pour la première fois, elle ne sera pas trouvée et donc elle sera construite (à l'exécution ultérieure, l'image construite sera trouvée et renvoyée).
L'image personnalisée est construite avec [_build_sandbox_image()](https://github.com/OpenDevin/OpenDevin/blob/main/opendevin/runtime/docker/image_agnostic_util.py#L29), qui crée un fichier docker en utilisant votre image personnalisée comme base et configure ensuite l'environnement pour OpenDevin, comme ceci:
L'image personnalisée est construite avec [_build_sandbox_image()](https://github.com/All-Hands-AI/OpenHands/blob/main/openhands/runtime/docker/image_agnostic_util.py#L29), qui crée un fichier docker en utilisant votre image personnalisée comme base et configure ensuite l'environnement pour OpenHands, comme ceci:
> Remarque: Le nom de l'image est modifié via [_get_new_image_name()](https://github.com/OpenDevin/OpenDevin/blob/main/opendevin/runtime/docker/image_agnostic_util.py#L63) et c'est ce nom modifié qui sera recherché lors des exécutions ultérieures.
> Remarque: Le nom de l'image est modifié via [_get_new_image_name()](https://github.com/All-Hands-AI/OpenHands/blob/main/openhands/runtime/docker/image_agnostic_util.py#L63) et c'est ce nom modifié qui sera recherché lors des exécutions ultérieures.
## Dépannage / Erreurs
### Erreur: ```useradd: UID 1000 est non unique```
Si vous voyez cette erreur dans la sortie de la console, il s'agit du fait que OpenDevin essaie de créer le utilisateur opendevin dans le sandbox avec un ID d'utilisateur de 1000, cependant cet ID d'utilisateur est déjà utilisé dans l'image (pour une raison inconnue). Pour résoudre ce problème, changez la valeur du champ sandbox_user_id dans le fichier config.toml en une valeur différente:
Si vous voyez cette erreur dans la sortie de la console, il s'agit du fait que OpenHands essaie de créer le utilisateur openhands dans le sandbox avec un ID d'utilisateur de 1000, cependant cet ID d'utilisateur est déjà utilisé dans l'image (pour une raison inconnue). Pour résoudre ce problème, changez la valeur du champ sandbox_user_id dans le fichier config.toml en une valeur différente:
@@ -104,4 +98,4 @@ Si vous voyez un message d'erreur indiquant que le port est utilisé ou indispon
## Discuter
Pour d'autres problèmes ou questions rejoignez le [Slack](https://join.slack.com/t/opendevin/shared_invite/zt-2jsrl32uf-fTeeFjNyNYxqSZt5NPY3fA) ou le [Discord](https://discord.gg/ESHStjSjD4) et demandez!
Pour d'autres problèmes ou questions rejoignez le [Slack](https://join.slack.com/t/opendevin/shared_invite/zt-2oikve2hu-UDxHeo8nsE69y6T7yFX_BA) ou le [Discord](https://discord.gg/ESHStjSjD4) et demandez!
Lorsque vous utilisez OpenDevin, vous rencontrerez sans aucun doute des cas où les choses fonctionnent bien et d'autres où elles ne fonctionnent pas. Nous vous encourageons à fournir des commentaires lorsque vous utilisez OpenDevin pour aider l'équipe de développement et, peut-être plus important encore, créer un corpus ouvert d'exemples de formation pour les agents de codage -- Partagez-OpenDevin !
# ✅ Fournir des commentaires
## 📝 Comment Fournir des Commentaires
Lorsque vous utilisez OpenHands, vous rencontrerez des cas où les choses fonctionnent bien, et d'autres où elles ne fonctionnent pas. Nous vous encourageons à fournir des commentaires lorsque vous utilisez OpenHands pour aider à donner des retours à l'équipe de développement, et peut-être plus important encore, créer un corpus ouvert d'exemples d'entraînement d'agents de codage -- Share-OpenHands !
Fournir des commentaires est simple ! Lorsque vous utilisez OpenDevin, vous pouvez appuyer sur le bouton de pouce vers le haut ou vers le bas à n'importe quel moment de votre interaction. Vous serez invité à fournir votre adresse email (par exemple, afin que nous puissions vous contacter si nous voulons poser des questions de suivi), et vous pouvez choisir si vous souhaitez fournir des commentaires publiquement ou en privé.
Fournir des commentaires est facile ! Lorsque vous utilisez OpenHands, vous pouvez appuyer sur le bouton pouce vers le haut ou pouce vers le bas à tout moment pendant votre interaction. Vous serez invité à fournir votre adresse e-mail (par exemple, afin que nous puissions vous contacter si nous voulons poser des questions de suivi), et vous pouvez choisir si vous souhaitez fournir des commentaires publiquement ou en privé.
* Les données **publiques** seront distribuées sous la licence MIT, comme OpenDevin lui-même, et pourront être utilisées par la communauté pour former et tester des modèles. Évidemment, les commentaires que vous pouvez rendre publics seront plus précieux pour la communauté dans son ensemble, donc lorsque vous ne traitez pas d'informations sensibles, nous vous encourageons à choisir cette option !
* Les données **privées** ne seront partagées qu'avec l'équipe OpenDevin dans le but d'améliorer OpenDevin.
## 📜 Utilisation des données et confidentialité
### Paramètres de partage des données
Lorsque vous soumettez des données, vous pouvez les soumettre publiquement ou en privé.
* Les données **publiques** seront distribuées sous la licence MIT, comme OpenHands lui-même, et pourront être utilisées par la communauté pour entraîner et tester des modèles. Évidemment, les commentaires que vous pouvez rendre publics seront plus précieux pour la communauté dans son ensemble, donc lorsque vous ne traitez pas d'informations sensibles, nous vous encourageons à choisir cette option !
* Les données **privées** ne seront partagées qu'avec l'équipe OpenHands dans le but d'améliorer OpenHands.
### Qui collecte et stocke les données ?
Les données sont collectées et stockées par [All Hands AI](https://all-hands.dev), une entreprise fondée par les mainteneurs d'OpenHands pour soutenir et améliorer OpenHands.
### Comment les données publiques seront-elles publiées ?
Les données publiques seront publiées lorsque nous atteindrons des jalons fixes, tels que 1 000 exemples publics, 10 000 exemples publics, etc.
À ce moment-là, nous suivrons le processus de publication suivant :
1. Toutes les personnes qui ont contribué à des commentaires publics recevront un e-mail décrivant la publication des données et auront la possibilité de se retirer.
2. La ou les personnes en charge de la publication des données effectueront un contrôle de la qualité des données, en supprimant les commentaires de mauvaise qualité, en supprimant les adresses e-mail des soumissionnaires et en essayant de supprimer toute information sensible.
3. Les données seront publiées publiquement sous la licence MIT via des sites couramment utilisés tels que GitHub ou Hugging Face.
### Que faire si je veux que mes données soient supprimées ?
Pour les données sur les serveurs d'All Hands AI, nous sommes heureux de les supprimer sur demande :
**Une pièce de données :** Si vous souhaitez supprimer une pièce de données, nous ajouterons prochainement un mécanisme pour supprimer les pièces de données en utilisant le lien et le mot de passe qui s'affichent sur l'interface lorsque vous soumettez des données.
**Toutes les données :** Si vous souhaitez que toutes vos données soient supprimées, ou si vous n'avez pas l'ID et le mot de passe que vous avez reçus lors de la soumission des données, veuillez contacter `contact@all-hands.dev` à partir de l'adresse e-mail que vous avez enregistrée lorsque vous avez initialement soumis les données.
OpenHands peut être exécuté en mode CLI interactif, ce qui permet aux utilisateurs de démarrer une session interactive via la ligne de commande.
Ce mode est différent du [mode headless](headless-mode), qui est non interactif et mieux adapté aux scripts.
## Avec Python
Pour démarrer une session OpenHands interactive via la ligne de commande, suivez ces étapes :
1. Assurez-vous d'avoir suivi les [instructions de configuration de développement](https://github.com/All-Hands-AI/OpenHands/blob/main/Development.md).
2. Exécutez la commande suivante :
```bash
poetry run python -m openhands.core.cli
```
Cette commande démarrera une session interactive où vous pourrez saisir des tâches et recevoir des réponses d'OpenHands.
Vous devrez vous assurer de définir votre modèle, votre clé API et d'autres paramètres via des variables d'environnement
[ou le fichier `config.toml`](https://github.com/All-Hands-AI/OpenHands/blob/main/config.template.toml).
## Avec Docker
Pour exécuter OpenHands en mode CLI avec Docker, suivez ces étapes :
1. Définissez `WORKSPACE_BASE` sur le répertoire que vous souhaitez qu'OpenHands modifie :
```bash
WORKSPACE_BASE=$(pwd)/workspace
```
2. Définissez `LLM_MODEL` sur le modèle que vous souhaitez utiliser :
```bash
LLM_MODEL="anthropic/claude-3-5-sonnet-20241022"
```
3. Définissez `LLM_API_KEY` sur votre clé API :
```bash
LLM_API_KEY="sk_test_12345"
```
4. Exécutez la commande Docker suivante :
```bash
docker run -it \
--pull=always \
-e SANDBOX_USER_ID=$(id -u)\
-e WORKSPACE_MOUNT_PATH=$WORKSPACE_BASE\
-e LLM_API_KEY=$LLM_API_KEY\
-e LLM_MODEL=$LLM_MODEL\
-v $WORKSPACE_BASE:/opt/workspace_base \
-v /var/run/docker.sock:/var/run/docker.sock \
--add-host host.docker.internal:host-gateway \
--name openhands-app-$(date +%Y%m%d%H%M%S)\
ghcr.io/all-hands-ai/openhands:0.11 \
python -m openhands.core.cli
```
Cette commande démarrera une session interactive dans Docker où vous pourrez saisir des tâches et recevoir des réponses d'OpenHands.
## Exemples de commandes CLI et de sorties attendues
Voici quelques exemples de commandes CLI et leurs sorties attendues :
### Exemple 1 : Tâche simple
```bash
Comment puis-je vous aider ? >> Écrivez un script Python qui affiche "Hello, World!"
```
Sortie attendue :
```bash
🤖 Bien sûr ! Voici un script Python qui affiche "Hello, World!" :
❯ print("Hello, World!")
```
### Exemple 2 : Commande Bash
```bash
Comment puis-je vous aider ? >> Créez un répertoire nommé "test_dir"
```
Sortie attendue :
```bash
🤖 Création d'un répertoire nommé "test_dir" :
❯ mkdir test_dir
```
### Exemple 3 : Gestion des erreurs
```bash
Comment puis-je vous aider ? >> Supprimez un fichier inexistant
Le sandbox est l'endroit où l'agent effectue ses tâches. Au lieu d'exécuter des commandes directement sur votre ordinateur (ce qui pourrait être risqué), l'agent les exécute à l'intérieur d'un conteneur Docker.
Le sandbox OpenHands par défaut (`python-nodejs:python3.12-nodejs22` de [nikolaik/python-nodejs](https://hub.docker.com/r/nikolaik/python-nodejs)) est livré avec certains paquets installés tels que Python et Node.js, mais peut nécessiter l'installation d'autres logiciels par défaut.
Vous avez deux options pour la personnalisation :
1. Utiliser une image existante avec les logiciels requis.
2. Créer votre propre image Docker personnalisée.
Si vous choisissez la première option, vous pouvez passer la section `Créer Votre Image Docker`.
## Créer Votre Image Docker
Pour créer une image Docker personnalisée, elle doit être basée sur Debian.
Par exemple, si vous voulez qu'OpenHands ait `ruby` installé, créez un `Dockerfile` avec le contenu suivant :
```dockerfile
FROMdebian:latest
# Installer les paquets requis
RUN apt-get update && apt-get install -y ruby
```
Enregistrez ce fichier dans un dossier. Ensuite, construisez votre image Docker (par exemple, nommée custom-image) en naviguant vers le dossier dans le terminal et en exécutant :
```bash
docker build -t custom-image .
```
Cela produira une nouvelle image appelée `custom-image`, qui sera disponible dans Docker.
> Notez que dans la configuration décrite dans ce document, OpenHands s'exécutera en tant qu'utilisateur "openhands" à l'intérieur du sandbox et donc tous les paquets installés via le docker file devraient être disponibles pour tous les utilisateurs du système, pas seulement root.
## Utiliser le Workflow de Développement
### Configuration
Tout d'abord, assurez-vous de pouvoir exécuter OpenHands en suivant les instructions dans [Development.md](https://github.com/All-Hands-AI/OpenHands/blob/main/Development.md).
### Spécifier l'Image de Base du Sandbox
Dans le fichier `config.toml` dans le répertoire OpenHands, définissez `sandbox_base_container_image` sur l'image que vous souhaitez utiliser. Cela peut être une image que vous avez déjà extraite ou une que vous avez construite :
```bash
[core]
...
sandbox_base_container_image="custom-image"
```
### Exécuter
Exécutez OpenHands en exécutant ```make run``` dans le répertoire de niveau supérieur.
## Explication Technique
Veuillez vous référer à la [section image docker personnalisée de la documentation d'exécution](https://docs.all-hands.dev/modules/usage/architecture/runtime#advanced-how-openhands-builds-and-maintains-od-runtime-images) pour plus de détails.
## Dépannage / Erreurs
### Erreur : ```useradd: UID 1000 is not unique```
Si vous voyez cette erreur dans la sortie de la console, c'est parce qu'OpenHands essaie de créer l'utilisateur openhands dans le sandbox avec un UID de 1000, mais cet UID est déjà utilisé dans l'image (pour une raison quelconque). Pour corriger cela, changez le champ sandbox_user_id dans le fichier config.toml à une valeur différente :
```toml
[core]
workspace_base="./workspace"
run_as_openhands=true
sandbox_base_container_image="custom_image"
sandbox_user_id="1001"
```
### Erreurs d'utilisation de port
Si vous voyez une erreur concernant un port déjà utilisé ou indisponible, essayez de supprimer tous les conteneurs Docker en cours d'exécution (exécutez `docker ps` et `docker rm` sur les conteneurs pertinents) puis réexécutez ```make run```.
## Discuter
Pour d'autres problèmes ou questions, rejoignez le [Slack](https://join.slack.com/t/opendevin/shared_invite/zt-2oikve2hu-UDxHeo8nsE69y6T7yFX_BA) ou le [Discord](https://discord.gg/ESHStjSjD4) et demandez !
Ce qui suit est destiné à servir d'introduction au débogage d'OpenHands à des fins de développement.
## Serveur / VSCode
Le `launch.json` suivant permettra de déboguer les éléments agent, contrôleur et serveur, mais pas le bac à sable (qui s'exécute dans docker). Il ignorera toutes les modifications à l'intérieur du répertoire `workspace/` :
```
{
"version": "0.2.0",
"configurations": [
{
"name": "OpenHands CLI",
"type": "debugpy",
"request": "launch",
"module": "openhands.core.cli",
"justMyCode": false
},
{
"name": "OpenHands WebApp",
"type": "debugpy",
"request": "launch",
"module": "uvicorn",
"args": [
"openhands.server.listen:app",
"--reload",
"--reload-exclude",
"${workspaceFolder}/workspace",
"--port",
"3000"
],
"justMyCode": false
}
]
}
```
Des configurations de débogage plus spécifiques qui incluent plus de paramètres peuvent être spécifiées :
```
...
{
"name": "Debug CodeAct",
"type": "debugpy",
"request": "launch",
"module": "openhands.core.main",
"args": [
"-t",
"Demandez-moi quelle est votre tâche.",
"-d",
"${workspaceFolder}/workspace",
"-c",
"CodeActAgent",
"-l",
"llm.o1",
"-n",
"prompts"
],
"justMyCode": false
}
...
```
Les valeurs dans l'extrait ci-dessus peuvent être mises à jour de telle sorte que :
* *t* : la tâche
* *d* : le répertoire de l'espace de travail openhands
* *c* : l'agent
* *l* : la configuration LLM (prédéfinie dans config.toml)
* *n* : le nom de la session (par exemple, le nom du flux d'événements)
Ce guide fournit un aperçu de la façon d'intégrer votre propre benchmark d'évaluation dans le framework OpenHands.
## Configuration de l'environnement et de la configuration LLM
Veuillez suivre les instructions [ici](https://github.com/All-Hands-AI/OpenHands/blob/main/Development.md) pour configurer votre environnement de développement local.
OpenHands en mode développement utilise `config.toml` pour garder une trace de la plupart des configurations.
Voici un exemple de fichier de configuration que vous pouvez utiliser pour définir et utiliser plusieurs LLMs :
```toml
[llm]
# IMPORTANT : ajoutez votre clé API ici et définissez le modèle que vous souhaitez évaluer
model="claude-3-5-sonnet-20241022"
api_key="sk-XXX"
[llm.eval_gpt4_1106_preview_llm]
model="gpt-4-1106-preview"
api_key="XXX"
temperature=0.0
[llm.eval_some_openai_compatible_model_llm]
model="openai/MODEL_NAME"
base_url="https://OPENAI_COMPATIBLE_URL/v1"
api_key="XXX"
temperature=0.0
```
## Comment utiliser OpenHands en ligne de commande
OpenHands peut être exécuté depuis la ligne de commande en utilisant le format suivant :
```bash
poetry run python ./openhands/core/main.py \
-i <max_iterations> \
-t "<task_description>"\
-c <agent_class> \
-l <llm_config>
```
Par exemple :
```bash
poetry run python ./openhands/core/main.py \
-i 10\
-t "Écrivez-moi un script bash qui affiche hello world."\
-c CodeActAgent \
-l llm
```
Cette commande exécute OpenHands avec :
- Un maximum de 10 itérations
- La description de tâche spécifiée
- En utilisant CodeActAgent
- Avec la configuration LLM définie dans la section `llm` de votre fichier `config.toml`
## Comment fonctionne OpenHands
Le point d'entrée principal d'OpenHands se trouve dans `openhands/core/main.py`. Voici un flux simplifié de son fonctionnement :
1. Analyse des arguments de ligne de commande et chargement de la configuration
2. Création d'un environnement d'exécution à l'aide de `create_runtime()`
3. Initialisation de l'agent spécifié
4. Exécution du contrôleur à l'aide de `run_controller()`, qui :
- Attache l'environnement d'exécution à l'agent
- Exécute la tâche de l'agent
- Renvoie un état final une fois terminé
La fonction `run_controller()` est le cœur de l'exécution d'OpenHands. Elle gère l'interaction entre l'agent, l'environnement d'exécution et la tâche, en gérant des choses comme la simulation d'entrée utilisateur et le traitement des événements.
## Le moyen le plus simple de commencer : Explorer les benchmarks existants
Nous vous encourageons à examiner les différents benchmarks d'évaluation disponibles dans le [répertoire `evaluation/`](https://github.com/All-Hands-AI/OpenHands/blob/main/evaluation) de notre dépôt.
Pour intégrer votre propre benchmark, nous vous suggérons de commencer par celui qui ressemble le plus à vos besoins. Cette approche peut considérablement rationaliser votre processus d'intégration, vous permettant de vous appuyer sur les structures existantes et de les adapter à vos exigences spécifiques.
## Comment créer un workflow d'évaluation
Pour créer un workflow d'évaluation pour votre benchmark, suivez ces étapes :
1. Importez les utilitaires OpenHands pertinents :
```python
import openhands.agenthub
from evaluation.utils.shared import (
EvalMetadata,
EvalOutput,
make_metadata,
prepare_dataset,
reset_logger_for_multiprocessing,
run_evaluation,
)
from openhands.controller.state.state import State
from openhands.core.config import (
AppConfig,
SandboxConfig,
get_llm_config_arg,
parse_arguments,
)
from openhands.core.logger import openhands_logger as logger
from openhands.core.main import create_runtime, run_controller
from openhands.events.action import CmdRunAction
from openhands.events.observation import CmdOutputObservation, ErrorObservation
Ce workflow configure la configuration, initialise l'environnement d'exécution, traite chaque instance en exécutant l'agent et en évaluant ses actions, puis collecte les résultats dans un objet `EvalOutput`. La fonction `run_evaluation` gère la parallélisation et le suivi de la progression.
N'oubliez pas de personnaliser les fonctions `get_instruction`, `your_user_response_function` et `evaluate_agent_actions` en fonction des exigences spécifiques de votre benchmark.
En suivant cette structure, vous pouvez créer un workflow d'évaluation robuste pour votre benchmark dans le framework OpenHands.
## Comprendre la `user_response_fn`
La `user_response_fn` est un composant crucial dans le workflow d'évaluation d'OpenHands. Elle simule l'interaction de l'utilisateur avec l'agent, permettant des réponses automatisées pendant le processus d'évaluation. Cette fonction est particulièrement utile lorsque vous souhaitez fournir des réponses cohérentes et prédéfinies aux requêtes ou actions de l'agent.
### Workflow et interaction
Le workflow correct pour gérer les actions et la `user_response_fn` est le suivant :
1. L'agent reçoit une tâche et commence à la traiter
2. L'agent émet une Action
3. Si l'Action est exécutable (par exemple, CmdRunAction, IPythonRunCellAction) :
- Le Runtime traite l'Action
- Le Runtime renvoie une Observation
4. Si l'Action n'est pas exécutable (généralement une MessageAction) :
- La `user_response_fn` est appelée
- Elle renvoie une réponse utilisateur simulée
5. L'agent reçoit soit l'Observation, soit la réponse simulée
6. Les étapes 2 à 5 se répètent jusqu'à ce que la tâche soit terminée ou que le nombre maximum d'itérations soit atteint
Voici une représentation visuelle plus précise :
```
[Agent]
|
v
[Émettre une Action]
|
v
[L'Action est-elle exécutable ?]
/ \
Oui Non
| |
v v
[Runtime] [user_response_fn]
| |
v v
[Renvoyer une Observation] [Réponse simulée]
\ /
\ /
v v
[L'agent reçoit le feedback]
|
v
[Continuer ou terminer la tâche]
```
Dans ce workflow :
- Les actions exécutables (comme l'exécution de commandes ou de code) sont gérées directement par le Runtime
- Les actions non exécutables (généralement lorsque l'agent veut communiquer ou demander des clarifications) sont gérées par la `user_response_fn`
- L'agent traite ensuite le feedback, qu'il s'agisse d'une Observation du Runtime ou d'une réponse simulée de la `user_response_fn`
Cette approche permet une gestion automatisée des actions concrètes et des interactions utilisateur simulées, ce qui la rend adaptée aux scénarios d'évaluation où vous souhaitez tester la capacité de l'agent à effectuer des tâches avec une intervention humaine minimale.
### Exemple d'implémentation
Voici un exemple de `user_response_fn` utilisée dans l'évaluation SWE-Bench :
```python
def codeact_user_response(state: State | None) -> str:
msg = (
'Veuillez continuer à travailler sur la tâche avec l\'approche que vous jugez appropriée.\n'
'Si vous pensez avoir résolu la tâche, veuillez d\'abord envoyer votre réponse à l\'utilisateur via un message, puis <execute_bash> exit </execute_bash>.\n'
'IMPORTANT : VOUS NE DEVEZ JAMAIS DEMANDER DE L\'AIDE HUMAINE.\n'
)
if state and state.history:
# vérifier si l'agent a essayé de parler à l'utilisateur 3 fois, si oui, faire savoir à l'agent qu'il peut abandonner
user_msgs = [
event
for event in state.history
if isinstance(event, MessageAction) and event.source == 'user'
]
if len(user_msgs) >= 2:
# faire savoir à l'agent qu'il peut abandonner lorsqu'il a essayé 3 fois
1. Fournit un message standard encourageant l'agent à continuer à travailler
2. Vérifie combien de fois l'agent a tenté de communiquer avec l'utilisateur
3. Si l'agent a fait plusieurs tentatives, il lui donne la possibilité d'abandonner
En utilisant cette fonction, vous pouvez garantir un comportement cohérent sur plusieurs exécutions d'évaluation et empêcher l'agent de rester bloqué en attendant une entrée humaine.
Ce guide explique comment utiliser l'Action GitHub OpenHands, à la fois dans le dépôt OpenHands et dans vos propres projets.
## Utilisation de l'Action dans le dépôt OpenHands
Pour utiliser l'Action GitHub OpenHands dans le dépôt OpenHands, un mainteneur OpenHands peut :
1. Créer une issue dans le dépôt.
2. Ajouter le label `fix-me` à l'issue.
3. L'action se déclenchera automatiquement et tentera de résoudre l'issue.
## Installation de l'Action dans un nouveau dépôt
Pour installer l'Action GitHub OpenHands dans votre propre dépôt, suivez les [instructions dans le dépôt OpenHands Resolver](https://github.com/All-Hands-AI/OpenHands-resolver?tab=readme-ov-file#using-the-github-actions-workflow).
OpenHands fournit un mode Interface Graphique (GUI) convivial pour interagir avec l'assistant IA. Ce mode offre une façon intuitive de configurer l'environnement, gérer les paramètres et communiquer avec l'IA.
## Installation et Configuration
1. Suivez les instructions du guide d'[Installation](../installation) pour installer OpenHands.
2. Après avoir exécuté la commande, accédez à OpenHands à l'adresse [http://localhost:3000](http://localhost:3000).
## Interagir avec l'Interface Graphique
### Configuration Initiale
1. Lors du premier lancement, vous verrez une fenêtre modale de paramètres.
2. Sélectionnez un `Fournisseur LLM` et un `Modèle LLM` dans les menus déroulants.
3. Entrez la `Clé API` correspondante pour le fournisseur choisi.
4. Cliquez sur "Enregistrer" pour appliquer les paramètres.
### Paramètres Avancés
1. Activez `Options Avancées` pour accéder aux paramètres supplémentaires.
2. Utilisez la zone de texte `Modèle Personnalisé` pour entrer manuellement un modèle s'il n'est pas dans la liste.
3. Spécifiez une `URL de Base` si requise par votre fournisseur LLM.
### Interface Principale
L'interface principale se compose de plusieurs éléments clés :
1.**Fenêtre de Chat** : La zone centrale où vous pouvez voir l'historique de conversation avec l'assistant IA.
2.**Zone de Saisie** : Située en bas de l'écran, utilisez-la pour taper vos messages ou commandes à l'IA.
3.**Bouton Envoyer** : Cliquez dessus pour envoyer votre message à l'IA.
4.**Bouton Paramètres** : Une icône d'engrenage qui ouvre la fenêtre modale des paramètres, vous permettant d'ajuster votre configuration à tout moment.
5.**Panneau Espace de Travail** : Affiche les fichiers et dossiers de votre espace de travail, vous permettant de naviguer et visualiser les fichiers, ou l'historique des commandes passées de l'agent ou de navigation web.
### Interagir avec l'IA
1. Tapez votre question, requête ou description de tâche dans la zone de saisie.
2. Cliquez sur le bouton envoyer ou appuyez sur Entrée pour soumettre votre message.
3. L'IA traitera votre saisie et fournira une réponse dans la fenêtre de chat.
4. Vous pouvez poursuivre la conversation en posant des questions de suivi ou en fournissant des informations supplémentaires.
## Conseils pour une Utilisation Efficace
1. Soyez spécifique dans vos requêtes pour obtenir les réponses les plus précises et utiles, comme décrit dans les [meilleures pratiques d'invite](../prompting-best-practices).
2. Utilisez le panneau d'espace de travail pour explorer la structure de votre projet.
3. Utilisez l'un des modèles recommandés, comme décrit dans la section [LLMs](usage/llms/llms.md).
N'oubliez pas, le mode Interface Graphique d'OpenHands est conçu pour rendre votre interaction avec l'assistant IA aussi fluide et intuitive que possible. N'hésitez pas à explorer ses fonctionnalités pour maximiser votre productivité.
Il existe différentes façons d'exécuter OpenHands sur Kubernetes ou OpenShift. Ce guide présente une façon possible :
1. Créer un PV "en tant qu'administrateur du cluster" pour mapper les données workspace_base et le répertoire docker au pod via le nœud worker
2. Créer un PVC pour pouvoir monter ces PV sur le pod
3. Créer un pod qui contient deux conteneurs : les conteneurs OpenHands et Sandbox
## Étapes détaillées pour l'exemple ci-dessus
> Remarque : Assurez-vous d'être connecté au cluster avec le compte approprié pour chaque étape. La création de PV nécessite un administrateur de cluster !
> Assurez-vous d'avoir les autorisations de lecture/écriture sur le hostPath utilisé ci-dessous (c'est-à-dire /tmp/workspace)
1. Créer le PV :
Le fichier yaml d'exemple ci-dessous peut être utilisé par un administrateur de cluster pour créer le PV.
- workspace-pv.yaml
```yamlfile
apiVersion: v1
kind: PersistentVolume
metadata:
name: workspace-pv
spec:
capacity:
storage: 2Gi
accessModes:
- ReadWriteOnce
persistentVolumeReclaimPolicy: Retain
hostPath:
path: /tmp/workspace
```
```bash
# appliquer le fichier yaml
$ oc create -f workspace-pv.yaml
persistentvolume/workspace-pv created
# vérifier :
$ oc get pv
NAME CAPACITY ACCESS MODES RECLAIM POLICY STATUS CLAIM STORAGECLASS REASON AGE
workspace-pv 2Gi RWO Retain Available 7m23s
```
- docker-pv.yaml
```yamlfile
apiVersion: v1
kind: PersistentVolume
metadata:
name: docker-pv
spec:
capacity:
storage: 2Gi
accessModes:
- ReadWriteOnce
persistentVolumeReclaimPolicy: Retain
hostPath:
path: /var/run/docker.sock
```
```bash
# appliquer le fichier yaml
$ oc create -f docker-pv.yaml
persistentvolume/docker-pv created
# vérifier :
oc get pv
NAME CAPACITY ACCESS MODES RECLAIM POLICY STATUS CLAIM STORAGECLASS REASON AGE
docker-pv 2Gi RWO Retain Available 6m55s
workspace-pv 2Gi RWO Retain Available 7m23s
```
2. Créer le PVC :
Exemple de fichier yaml PVC ci-dessous :
- workspace-pvc.yaml
```yamlfile
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: workspace-pvc
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 1Gi
```
```bash
# créer le pvc
$ oc create -f workspace-pvc.yaml
persistentvolumeclaim/workspace-pvc created
# vérifier
$ oc get pvc
NAME STATUS VOLUME CAPACITY ACCESS MODES STORAGECLASS AGE
workspace-pvc Pending hcloud-volumes 4s
$ oc get events
LAST SEEN TYPE REASON OBJECT MESSAGE
8s Normal WaitForFirstConsumer persistentvolumeclaim/workspace-pvc waiting for first consumer to be created before binding
```
- docker-pvc.yaml
```yamlfile
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: docker-pvc
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 1Gi
```
```bash
# créer le pvc
$ oc create -f docker-pvc.yaml
persistentvolumeclaim/docker-pvc created
# vérifier
$ oc get pvc
NAME STATUS VOLUME CAPACITY ACCESS MODES STORAGECLASS AGE
docker-pvc Pending hcloud-volumes 4s
workspace-pvc Pending hcloud-volumes 2m53s
$ oc get events
LAST SEEN TYPE REASON OBJECT MESSAGE
10s Normal WaitForFirstConsumer persistentvolumeclaim/docker-pvc waiting for first consumer to be created before binding
10s Normal WaitForFirstConsumer persistentvolumeclaim/workspace-pvc waiting for first consumer to be created before binding
W0716 11:22:07.776271 107626 warnings.go:70] would violate PodSecurity "restricted:v1.24": allowPrivilegeEscalation !=false(containers "openhands-app-2024", "openhands-sandbox-2024" must set securityContext.allowPrivilegeEscalation=false), unrestricted capabilities (containers "openhands-app-2024", "openhands-sandbox-2024" must set securityContext.capabilities.drop=["ALL"]), runAsNonRoot !=true(pod or containers "openhands-app-2024", "openhands-sandbox-2024" must set securityContext.runAsNonRoot=true), seccompProfile (pod or containers "openhands-app-2024", "openhands-sandbox-2024" must set securityContext.seccompProfile.type to "RuntimeDefault" or "Localhost")
pod/openhands-app-2024 created
# L'avertissement ci-dessus peut être ignoré pour l'instant car nous ne modifierons pas les restrictions SCC.
# vérifier
$ oc get pods
NAME READY STATUS RESTARTS AGE
openhands-app-2024 0/2 Pending 0 5s
$ oc get pods
NAME READY STATUS RESTARTS AGE
openhands-app-2024 0/2 ContainerCreating 0 15s
$ oc get events
LAST SEEN TYPE REASON OBJECT MESSAGE
38s Normal WaitForFirstConsumer persistentvolumeclaim/docker-pvc waiting for first consumer to be created before binding
23s Normal ExternalProvisioning persistentvolumeclaim/docker-pvc waiting for a volume to be created, either by external provisioner "csi.hetzner.cloud" or manually created by system administrator
27s Normal Provisioning persistentvolumeclaim/docker-pvc External provisioner is provisioning volume for claim "openhands/docker-pvc"
17s Normal ProvisioningSucceeded persistentvolumeclaim/docker-pvc Successfully provisioned volume pvc-2b1d223a-1c8f-4990-8e3d-68061a9ae252
16s Normal Scheduled pod/openhands-app-2024 Successfully assigned All-Hands-AI/OpenHands-app-2024 to worker1.hub.internal.blakane.com
9s Normal SuccessfulAttachVolume pod/openhands-app-2024 AttachVolume.Attach succeeded for volume "pvc-2b1d223a-1c8f-4990-8e3d-68061a9ae252"
9s Normal SuccessfulAttachVolume pod/openhands-app-2024 AttachVolume.Attach succeeded for volume "pvc-31f15b25-faad-4665-a25f-201a530379af"
6s Normal AddedInterface pod/openhands-app-2024 Add eth0 [10.128.2.48/23] from openshift-sdn
6s Normal Pulled pod/openhands-app-2024 Container image "ghcr.io/all-hands-ai/openhands:main" already present on machine
6s Normal Created pod/openhands-app-2024 Created container openhands-app-2024
6s Normal Started pod/openhands-app-2024 Started container openhands-app-2024
6s Normal Pulled pod/openhands-app-2024 Container image "ghcr.io/all-hands-ai/sandbox:main" already present on machine
5s Normal Created pod/openhands-app-2024 Created container openhands-sandbox-2024
5s Normal Started pod/openhands-app-2024 Started container openhands-sandbox-2024
83s Normal WaitForFirstConsumer persistentvolumeclaim/workspace-pvc waiting for first consumer to be created before binding
27s Normal Provisioning persistentvolumeclaim/workspace-pvc External provisioner is provisioning volume for claim "openhands/workspace-pvc"
17s Normal ProvisioningSucceeded persistentvolumeclaim/workspace-pvc Successfully provisioned volume pvc-31f15b25-faad-4665-a25f-201a530379af
$ oc get pods
NAME READY STATUS RESTARTS AGE
openhands-app-2024 2/2 Running 0 23s
$ oc get pvc
NAME STATUS VOLUME CAPACITY ACCESS MODES STORAGECLASS AGE
**Avertissement** : ce déploiement accorde à l'application OpenHands l'accès au socket docker de Kubernetes, ce qui crée un risque de sécurité. Utilisez à vos propres risques.
1- Créer une politique pour l'accès privilégié
2- Créer des informations d'identification gke (facultatif)
3- Créer le déploiement openhands
4- Commandes de vérification et d'accès à l'interface utilisateur
5- Dépanner le pod pour vérifier le conteneur interne
Vous pouvez également exécuter OpenHands en mode [headless scriptable](https://docs.all-hands.dev/modules/usage/how-to/headless-mode), comme un [CLI interactif](https://docs.all-hands.dev/modules/usage/how-to/cli-mode), ou en utilisant l'[Action GitHub OpenHands](https://docs.all-hands.dev/modules/usage/how-to/github-action).
## Configuration
Après avoir exécuté la commande ci-dessus, vous trouverez OpenHands en cours d'exécution à l'adresse [http://localhost:3000](http://localhost:3000).
L'agent aura accès au dossier `./workspace` pour effectuer son travail. Vous pouvez copier du code existant ici, ou modifier `WORKSPACE_BASE` dans la
commande pour pointer vers un dossier existant.
Au lancement d'OpenHands, vous verrez une fenêtre modale de paramètres. Vous **devez** sélectionner un `Fournisseur LLM` et un `Modèle LLM` et entrer une `Clé API` correspondante.
Ceux-ci peuvent être modifiés à tout moment en sélectionnant le bouton `Paramètres` (icône d'engrenage) dans l'interface utilisateur.
Si le `Modèle LLM` requis n'existe pas dans la liste, vous pouvez activer les `Options avancées` et le saisir manuellement avec le préfixe correct
dans la zone de texte `Modèle personnalisé`.
Les `Options avancées` vous permettent également de spécifier une `URL de base` si nécessaire.
La commande ci-dessus récupère la version stable la plus récente d'OpenHands. Vous avez également d'autres options :
- Pour une version spécifique, utilisez `ghcr.io/all-hands-ai/openhands:$VERSION`, en remplaçant $VERSION par le numéro de version.
- Nous utilisons semver et publions des tags majeurs, mineurs et de correctifs. Ainsi, `0.9` pointera automatiquement vers la dernière version `0.9.x`, et `0` pointera vers la dernière version `0.x.x`.
- Pour la version de développement la plus à jour, vous pouvez utiliser `ghcr.io/all-hands-ai/openhands:main`. Cette version est instable et n'est recommandée qu'à des fins de test ou de développement.
Vous pouvez choisir le tag qui correspond le mieux à vos besoins en fonction des exigences de stabilité et des fonctionnalités souhaitées.
Pour le workflow de développement, consultez [Development.md](https://github.com/All-Hands-AI/OpenHands/blob/main/Development.md).
Vous rencontrez des problèmes ? Consultez notre [Guide de dépannage](https://docs.all-hands.dev/modules/usage/troubleshooting).
OpenDevin est un **ingénieur logiciel IA autonome** capable d'exécuter des tâches d'ingénierie complexes et de collaborer activement avec les utilisateurs sur des projets de développement logiciel.
OpenHands est un **ingénieur logiciel IA autonome** capable d'exécuter des tâches d'ingénierie complexes et de collaborer activement avec les utilisateurs sur des projets de développement logiciel.
Ce projet est entièrement open-source, vous pouvez donc l'utiliser et le modifier comme bon vous semble.
:::tip
Explorez le code source d'OpenDevin sur [GitHub](https://github.com/OpenDevin/OpenDevin) ou rejoignez l'une de nos communautés !
Explorez le code source d'OpenHands sur [GitHub](https://github.com/All-Hands-AI/OpenHands) ou rejoignez l'une de nos communautés !
@@ -58,10 +58,10 @@ Explorez le code source d'OpenDevin sur [GitHub](https://github.com/OpenDevin/Op
## 🛠️ Pour commencer
La manière la plus simple d'exécuter OpenDevin est à l'intérieur d'un conteneur Docker. Il fonctionne mieux avec la version la plus récente de Docker, `26.0.0`.
La manière la plus simple d'exécuter OpenHands est à l'intérieur d'un conteneur Docker. Il fonctionne mieux avec la version la plus récente de Docker, `26.0.0`.
Vous devez utiliser Linux, Mac OS ou WSL sur Windows.
Pour démarrer OpenDevin dans un conteneur docker, exécutez les commandes suivantes dans votre terminal :
Pour démarrer OpenHands dans un conteneur docker, exécutez les commandes suivantes dans votre terminal :
:::warning
Lorsque vous exécutez la commande suivante, les fichiers dans `./workspace` peuvent être modifiés ou supprimés.
Vous trouverez OpenDevin fonctionnant à l'adresse [http://localhost:3000](http://localhost:3000) avec accès à `./workspace`. Pour qu'OpenDevin fonctionne sur votre code, placez-le dans `./workspace`.
Vous trouverez OpenHands fonctionnant à l'adresse [http://localhost:3000](http://localhost:3000) avec accès à `./workspace`. Pour qu'OpenHands fonctionne sur votre code, placez-le dans `./workspace`.
OpenDevin n'aura accès qu'à ce dossier de workspace. Le reste de votre système ne sera pas affecté car il s'exécute dans un bac à sable sécurisé de docker.
OpenHands n'aura accès qu'à ce dossier de workspace. Le reste de votre système ne sera pas affecté car il s'exécute dans un bac à sable sécurisé de docker.
:::tip
Si vous souhaitez utiliser la version **(instable !)** la plus récente, vous pouvez utiliser `ghcr.io/opendevin/opendevin:main` comme image (dernière ligne).
Si vous souhaitez utiliser la version **(instable !)** la plus récente, vous pouvez utiliser `ghcr.io/all-hands-ai/openhands:main` comme image (dernière ligne).
:::
Pour le workflow de développement, consultez [Development.md](https://github.com/OpenDevin/OpenDevin/blob/main/Development.md).
Pour le workflow de développement, consultez [Development.md](https://github.com/All-Hands-AI/OpenHands/blob/main/Development.md).
Avez-vous des problèmes ? Consultez notre [Guide de dépannage](https://docs.all-hands.dev/modules/usage/troubleshooting).
:::warning
OpenDevin est actuellement en cours de développement, mais vous pouvez déjà exécuter la version alpha pour voir le système de bout en bout en action.
OpenHands est actuellement en cours de développement, mais vous pouvez déjà exécuter la version alpha pour voir le système de bout en bout en action.
OpenHands utilise LiteLLM pour faire des appels aux modèles de chat d'Azure. Vous pouvez trouver leur documentation sur l'utilisation d'Azure comme fournisseur [ici](https://docs.litellm.ai/docs/providers/azure).
## Configuration d'Azure OpenAI
Lorsque vous exécutez OpenHands, vous devrez définir la variable d'environnement suivante en utilisant `-e` dans la
LLM_API_VERSION="<api-version>" # par exemple "2023-05-15"
```
Exemple :
```bash
docker run -it --pull=always \
-e LLM_API_VERSION="2023-05-15"
...
```
Ensuite, définissez les éléments suivants dans l'interface utilisateur d'OpenHands via les paramètres :
:::note
Vous aurez besoin du nom de votre déploiement ChatGPT qui peut être trouvé sur la page des déploiements dans Azure. Il est référencé comme
<deployment-name> ci-dessous.
:::
* Activez `Advanced Options`
*`Custom Model` à azure/<deployment-name>
*`Base URL` à votre URL de base de l'API Azure (par exemple `https://example-endpoint.openai.azure.com`)
*`API Key` à votre clé API Azure
## Embeddings
OpenHands utilise llama-index pour les embeddings. Vous pouvez trouver leur documentation sur Azure [ici](https://docs.llamaindex.ai/en/stable/api_reference/embeddings/azure_openai/).
### Configuration d'Azure OpenAI
Lorsque vous exécutez OpenHands, définissez les variables d'environnement suivantes en utilisant `-e` dans la
OpenDevin utilise LiteLLM pour les appels de complétion. Vous pouvez trouver leur documentation sur Azure [ici](https://docs.litellm.ai/docs/providers/azure)
OpenHands utilise LiteLLM pour les appels de complétion. Vous pouvez trouver leur documentation sur Azure [ici](https://docs.litellm.ai/docs/providers/azure)
### Configurations openai Azure
Lors de l'exécution de l'image Docker OpenDevin, vous devrez définir les variables d'environnement suivantes en utilisant `-e` :
Lors de l'exécution de l'image Docker OpenHands, vous devrez définir les variables d'environnement suivantes en utilisant `-e` :
```
LLM_BASE_URL="<azure-api-base-url>" # e.g. "https://openai-gpt-4-test-v-1.openai.azure.com/"
@@ -16,19 +16,19 @@ LLM_API_VERSION = "<api-version>" # e.g. "2024-02-15-preview"
```
:::note
Vous pouvez trouver le nom de votre déploiement ChatGPT sur la page des déploiements sur Azure. Par défaut ou initialement, il pourrait être le même que le nom du modèle de chat (par exemple 'GPT4-1106-preview'), mais il n'est pas obligé de l'être. Exécutez OpenDevin, et une fois chargé dans le navigateur, allez dans Paramètres et définissez le modèle comme suit : "azure/<your-actual-gpt-deployment-name>". Si ce n'est pas dans la liste, entrez votre propre texte et enregistrez-le.
Vous pouvez trouver le nom de votre déploiement ChatGPT sur la page des déploiements sur Azure. Par défaut ou initialement, il pourrait être le même que le nom du modèle de chat (par exemple 'GPT4-1106-preview'), mais il n'est pas obligé de l'être. Exécutez OpenHands, et une fois chargé dans le navigateur, allez dans Paramètres et définissez le modèle comme suit : "azure/<your-actual-gpt-deployment-name>". Si ce n'est pas dans la liste, entrez votre propre texte et enregistrez-le.
:::
## Embeddings
OpenDevin utilise llama-index pour les embeddings. Vous pouvez trouver leur documentation sur Azure [ici](https://docs.llamaindex.ai/en/stable/api_reference/embeddings/azure_openai/)
OpenHands utilise llama-index pour les embeddings. Vous pouvez trouver leur documentation sur Azure [ici](https://docs.llamaindex.ai/en/stable/api_reference/embeddings/azure_openai/)
### Configurations openai Azure
Le modèle utilisé pour les embeddings Azure OpenAI est "text-embedding-ada-002".
Vous avez besoin du nom de déploiement correct pour ce modèle dans votre compte Azure.
Lors de l'exécution d'OpenDevin dans Docker, définissez les variables d'environnement suivantes en utilisant `-e` :
Lors de l'exécution d'OpenHands dans Docker, définissez les variables d'environnement suivantes en utilisant `-e` :
OpenHands utilise LiteLLM pour faire des appels aux modèles de chat de Google. Vous pouvez trouver leur documentation sur l'utilisation de Google comme fournisseur :
- [Gemini - Google AI Studio](https://docs.litellm.ai/docs/providers/gemini)
- [VertexAI - Google Cloud Platform](https://docs.litellm.ai/docs/providers/vertex)
## Configurations de Gemini - Google AI Studio
Lors de l'exécution d'OpenHands, vous devrez définir les éléments suivants dans l'interface utilisateur d'OpenHands via les paramètres :
*`LLM Provider` à `Gemini`
*`LLM Model` au modèle que vous utiliserez.
Si le modèle ne figure pas dans la liste, activez `Advanced Options` et entrez-le dans `Custom Model` (par exemple, gemini/<model-name> comme `gemini/gemini-1.5-pro`).
*`API Key` à votre clé API Gemini
## Configurations de VertexAI - Google Cloud Platform
Pour utiliser Vertex AI via Google Cloud Platform lors de l'exécution d'OpenHands, vous devrez définir les variables d'environnement suivantes en utilisant `-e` dans la [commande docker run](/modules/usage/installation#start-the-app) :
OpenDevin utilise LiteLLM pour les appels de complétion. Les ressources suivantes sont pertinentes pour utiliser OpenDevin avec les LLMs de Google :
OpenHands utilise LiteLLM pour les appels de complétion. Les ressources suivantes sont pertinentes pour utiliser OpenHands avec les LLMs de Google :
- [Gemini - Google AI Studio](https://docs.litellm.ai/docs/providers/gemini)
- [VertexAI - Google Cloud Platform](https://docs.litellm.ai/docs/providers/vertex)
### Configurations de Gemini - Google AI Studio
Pour utiliser Gemini via Google AI Studio lors de l'exécution de l'image Docker d'OpenDevin, vous devez définir les variables d'environnement suivantes en utilisant `-e` :
Pour utiliser Gemini via Google AI Studio lors de l'exécution de l'image Docker d'OpenHands, vous devez définir les variables d'environnement suivantes en utilisant `-e` :
### Configurations de Vertex AI - Google Cloud Platform
Pour utiliser Vertex AI via Google Cloud Platform lors de l'exécution de l'image Docker d'OpenDevin, vous devez définir les variables d'environnement suivantes en utilisant `-e` :
Pour utiliser Vertex AI via Google Cloud Platform lors de l'exécution de l'image Docker d'OpenHands, vous devez définir les variables d'environnement suivantes en utilisant `-e` :
OpenHands utilise LiteLLM pour faire des appels aux modèles de chat sur Groq. Vous pouvez trouver leur documentation sur l'utilisation de Groq comme fournisseur [ici](https://docs.litellm.ai/docs/providers/groq).
## Configuration
Lorsque vous exécutez OpenHands, vous devrez définir les éléments suivants dans l'interface utilisateur d'OpenHands via les paramètres :
*`LLM Provider` à `Groq`
*`LLM Model` au modèle que vous utiliserez. [Visitez ici pour voir la liste des modèles hébergés par Groq](https://console.groq.com/docs/models). Si le modèle n'est pas dans la liste, activez les `Advanced Options` et entrez-le dans `Custom Model` (par exemple, groq/<model-name> comme `groq/llama3-70b-8192`).
*`API key` à votre clé API Groq. Pour trouver ou créer votre clé API Groq, [voir ici](https://console.groq.com/keys).
## Utilisation de Groq comme point de terminaison compatible OpenAI
Le point de terminaison Groq pour la complétion de chat est [principalement compatible OpenAI](https://console.groq.com/docs/openai). Par conséquent, vous pouvez accéder aux modèles Groq comme vous le feriez pour n'importe quel point de terminaison compatible OpenAI. Vous pouvez définir les éléments suivants dans l'interface utilisateur d'OpenHands via les paramètres :
* Activer les `Advanced Options`
*`Custom Model` au préfixe `openai/` + le modèle que vous utiliserez (par exemple, `openai/llama3-70b-8192`)
OpenHands peut se connecter à n'importe quel LLM pris en charge par LiteLLM. Cependant, il nécessite un modèle puissant pour fonctionner.
## Recommandations de modèles
Sur la base d'une évaluation récente des modèles de langage pour les tâches de codage (en utilisant le jeu de données SWE-bench), nous pouvons fournir quelques recommandations pour la sélection des modèles. L'analyse complète se trouve dans [cet article de blog](https://www.all-hands.dev/blog/evaluation-of-llms-as-coding-agents-on-swe-bench-at-30x-speed).
Lors du choix d'un modèle, tenez compte à la fois de la qualité des sorties et des coûts associés. Voici un résumé des résultats :
- Claude 3.5 Sonnet est le meilleur d'une bonne marge, atteignant un taux de résolution de 27% avec l'agent par défaut dans OpenHands.
- GPT-4o est à la traîne, et o1-mini a en fait obtenu des résultats légèrement inférieurs à ceux de GPT-4o. Nous avons analysé les résultats un peu, et brièvement, il semblait que o1 "réfléchissait trop" parfois, effectuant des tâches de configuration d'environnement supplémentaires alors qu'il aurait pu simplement aller de l'avant et terminer la tâche.
- Enfin, les modèles ouverts les plus puissants étaient Llama 3.1 405 B et deepseek-v2.5, et ils ont obtenu des résultats raisonnables, surpassant même certains des modèles fermés.
Veuillez vous référer à [l'article complet](https://www.all-hands.dev/blog/evaluation-of-llms-as-coding-agents-on-swe-bench-at-30x-speed) pour plus de détails.
Sur la base de ces résultats et des commentaires de la communauté, il a été vérifié que les modèles suivants fonctionnent raisonnablement bien avec OpenHands :
- claude-3-5-sonnet (recommandé)
- gpt-4 / gpt-4o
- llama-3.1-405b
- deepseek-v2.5
:::warning
OpenDevin émettra de nombreuses invitations au LLM que vous configurez. La plupart de ces LLM coûtent de l'argent -- assurez-vous de définir des limites de dépenses et de surveiller l'utilisation.
OpenHands enverra de nombreuses invites au LLM que vous configurez. La plupart de ces LLM sont payants, alors assurez-vous de définir des limites de dépenses et de surveiller l'utilisation.
:::
La variable d'environnement `LLM_MODEL` contrôle le modèle utilisé dans les interactions programmatiques.
Mais en utilisant l'interface utilisateur OpenDevin, vous devrez choisir votre modèle dans la fenêtre des paramètres (la roue dentée en bas à gauche).
Si vous avez réussi à exécuter OpenHands avec des LLM spécifiques qui ne figurent pas dans la liste, veuillez les ajouter à la liste vérifiée. Nous vous encourageons également à ouvrir une PR pour partager votre processus de configuration afin d'aider les autres utilisant le même fournisseur et LLM !
Les variables d'environnement suivantes peuvent être nécessaires pour certains LLM :
Pour une liste complète des fournisseurs et des modèles disponibles, veuillez consulter la [documentation litellm](https://docs.litellm.ai/docs/providers).
-`LLM_API_KEY`
-`LLM_BASE_URL`
:::note
La plupart des modèles locaux et open source actuels ne sont pas aussi puissants. Lorsque vous utilisez de tels modèles, vous pouvez constater de longs temps d'attente entre les messages, des réponses médiocres ou des erreurs concernant du JSON malformé. OpenHands ne peut être aussi puissant que les modèles qui le pilotent. Cependant, si vous en trouvez qui fonctionnent, veuillez les ajouter à la liste vérifiée ci-dessus.
:::
## Configuration LLM
Les éléments suivants peuvent être définis dans l'interface utilisateur d'OpenHands via les paramètres :
-`Fournisseur LLM`
-`Modèle LLM`
-`Clé API`
-`URL de base` (via `Paramètres avancés`)
Certains paramètres peuvent être nécessaires pour certains LLM/fournisseurs qui ne peuvent pas être définis via l'interface utilisateur. Au lieu de cela, ceux-ci peuvent être définis via des variables d'environnement passées à la [commande docker run](/modules/usage/installation#start-the-app) en utilisant `-e` :
-`LLM_API_VERSION`
-`LLM_EMBEDDING_MODEL`
-`LLM_EMBEDDING_DEPLOYMENT_NAME`
-`LLM_API_VERSION`
-`LLM_DROP_PARAMS`
-`LLM_DISABLE_VISION`
-`LLM_CACHING_PROMPT`
Nous avons quelques guides pour exécuter OpenDevin avec des fournisseurs de modèles spécifiques :
Nous avons quelques guides pour exécuter OpenHands avec des fournisseurs de modèles spécifiques :
- [ollama](llms/localLLMs)
- [Azure](llms/azureLLMs)
- [Azure](llms/azure-llms)
- [Google](llms/google-llms)
- [Groq](llms/groq)
- [OpenAI](llms/openai-llms)
- [OpenRouter](llms/openrouter)
Si vous utilisez un autre fournisseur, nous vous encourageons à ouvrir une PR pour partager votre configuration !
### Nouvelles tentatives d'API et limites de débit
## Remarque sur les modèles alternatifs
Les fournisseurs de LLM ont généralement des limites de débit, parfois très basses, et peuvent nécessiter de nouvelles tentatives. OpenHands réessaiera automatiquement les requêtes s'il reçoit une erreur de limite de débit (code d'erreur 429), une erreur de connexion API ou d'autres erreurs transitoires.
Les meilleurs modèles sont GPT-4 et Claude 3. Les modèles locaux et open source actuels ne sont pas aussi puissants.
Lors de l'utilisation d'un modèle alternatif, vous pouvez constater des temps d'attente prolongés entre les messages,
des réponses de mauvaise qualité ou des erreurs sur des JSON mal formés. OpenDevin
ne peut être aussi puissant que les modèles qui le pilotent -- heureusement, les membres de notre équipe travaillent activement à la construction de meilleurs modèles open source !
Vous pouvez personnaliser ces options selon vos besoins pour le fournisseur que vous utilisez. Consultez leur documentation et définissez les variables d'environnement suivantes pour contrôler le nombre de nouvelles tentatives et le temps entre les tentatives :
## Réessais d'API et limites de taux
-`LLM_NUM_RETRIES` (Par défaut 8)
-`LLM_RETRY_MIN_WAIT` (Par défaut 15 secondes)
-`LLM_RETRY_MAX_WAIT` (Par défaut 120 secondes)
-`LLM_RETRY_MULTIPLIER` (Par défaut 2)
Certains LLM ont des limites de taux et peuvent nécessiter des réessais. OpenDevin réessaiera automatiquement les demandes s'il reçoit une erreur 429 ou une erreur de connexion API.
Vous pouvez définir les variables d'environnement `LLM_NUM_RETRIES`, `LLM_RETRY_MIN_WAIT`, `LLM_RETRY_MAX_WAIT` pour contrôler le nombre de réessais et le temps entre les réessais.
Par défaut, `LLM_NUM_RETRIES` est 5 et `LLM_RETRY_MIN_WAIT`, `LLM_RETRY_MAX_WAIT` sont respectivement de 3 secondes et 60 secondes.
Si vous exécutez OpenHands en mode développement, vous pouvez également définir ces options dans le fichier `config.toml` :
```toml
[llm]
num_retries=8
retry_min_wait=15
retry_max_wait=120
retry_multiplier=2
```
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.