### Background
We need a way to set an execution quota per user for each block execution.
### Changes 🏗️
* Introduced a `UserBlockCredit`, a transaction table tracking the block usage along with it cost/quota.
* The tracking is toggled by `ENABLE_CREDIT` config, default = false.
* Introduced `BLOCK_COSTS` | `GET /blocks/costs` as a source of information for the cost on each block depending on the input configuration.
Improvements:
* Refactor logging in manager.py to always print a prefix and pass the metadata.
* Make executionStatus on AgentNodeExecution prisma enum. And add executionStatus on AgentGraphExecution.
* Use executionStatus from AgentGraphExecution to improve waiting logic on test_manager.py.
- feat(builder): Add "Stop Run" buttons to monitor and builder
- Implement additional state management in `useAgentGraph` hook
- Add "stop" request mechanism
- Implement execution status tracking using WebSockets
- Add `isSaving`, `isRunning`, `isStopping` outputs
- Add `requestStopRun` method
- Rename `requestSaveRun` to `requestSaveAndRun` for clarity
- Add needed functionality for the above to `AutoGPTServerAPI` client
- Add `stopGraphExecution` method
- Add support for multiple handlers per WebSocket method
- Fix parsing of timestamps in `execution_event` WebSocket messages
- Add `IconSquare` from Lucide to `@/components/ui/icons`
- feat(server): Add `POST /graphs/{graph_id}/executions/{graph_exec_id}/stop` route
- Add `stop_graph_run` method to `AgentServer`
- feat(server): Add `cancel_execution` method to `ExecutionManager`
- Replace node executor `ProcessPoolExecutor` by `multiprocessing.Pool` (which has a `terminate()` method)
- Remove now unnecessary `Executor.wait_future(..)` method
- Add `get_graph_execution(..)` in `.data.execution`
- fix(server): Reduce number of node executors to 5 per graph executor
This is necessary because `multiprocessing.Pool` spawns its workers on init, instead of based on demand like `ProcessPoolExecutor` does
- dx(server): Improve debug logging in `ExecutionManager`
- ci(server): Add debug logging mode to CI Pytest step
### Other improvements
Server:
- Improve output type of `ExecutionManager.add_execution(..)`
- Renamed a few things in `.server.rest_api` for consistency
Front end:
- Improved typing in `AutoGPTServerAPI` client
### Background
Currently, there is no way to construct the output of nodes into a composite data structure (list/dict/object) using the builder UI.
The backend already supports this feature by connecting the output pin to the input pin using these format:
* <pin_name>_$_<list_index> for constructing list
* <pin_name>_#_<dict_key> for constructing dict
* <pin_name>_@_<field_name> for constructing object
The scope of this PR is implementing the UX for this in the builder UI.
### Changes 🏗️
<img width="765" alt="image" src="https://github.com/user-attachments/assets/8fc319a4-1350-410f-98cf-24f2aa2bc34b">
This allows you to add more pins in a key value & list input: `_$_` list constructor & `_#_` dict constructor.
### Background
We need an explicit block for providing input & output for the graph.
This will later allow us to build a subgraph with pre-declared input & output schema.
This will also allow us to set input for the node in the middle of the graph, and enable a graph to have output values.
### Changes 🏗️
* Add InputBlock & OutputBlock
* Add graph structure validation on the graph execution step that asserts the following property:
- All mandatory input pin, has to be connected or have a default value, except the `InputBlock` node.
- All links have to connect valid nodes, and the sink & source name using the valid block field.
### Background
Pyro for API Server is not using Prisma, but still holding a Prisma connection.
The fast-API thread is also holding a Prisma connection, making Prisma connected in two different loop within a single process.
### Changes 🏗️
Disable a Prisma connection on Pyro thread for Server API process.
Fix test flakiness issue due to concurrency issue.
### Background
Input from the input pin is consumed only once. While this is required in most of the use cases, there are some cases where the input can only be produced once, and that input needs to be re-used just like an input default value, that is passively providing input data, without triggering any execution. The scope of this change is providing that functionality in the link level, this property will be called **`static link`** in this system.
### Changes 🏗️
Provides a static link feature with the following behaviours:
* A link can be marked `static` to become a static link.
* Once a node produces an output it will persist the output data and propagate the output to the other nodes through the link, for a static link, instead of making the data queued in the input pin, it will override the default value.
* Any input executions still waiting for the input will be backfilled using this output produced by the static link.
* And any upcoming executions that will use the input will always reuse the output produced by the static link.
See the added test to see the expected usage.
### Background
When multiple executors are executing the same node within the same graph execution, two node executions can read the same queue of input and read the same value—making the data that is supposed to be consumed once, consumed by two executions. The lack of lock & concurrency support for parallel execution within a single graph causes this issue.
Node concurrency also introduces poor UX in the current frontend implementation, when two nodes are executed in parallel, the current UI will not display its parallel execution update, but instead, it shows the updates that override each other. Until the execution observability is improved on the builder UI, this capability will be limited.
### Changes 🏗️
The scope of this change is to solve this issue by:
* Decouple Graph execution & Node execution, each has its own configured process pool.
* Make sure there is only 1 execution per node (we still allow parallel executions on different nodes) in a graph.
* Fixed concurrency issue by adding distributed lock API on agent_server.
* Few cleanups:
- Add more logging with geid & neid prefix on graph/node executions
- Moved execution status update to agent-server for a single source of status update (required by conn-manager/web-socket)
- Configured node parallelism to 10 & graph parallelism to 10 by default, so in the very rare worst-case, there can be 100 node executions.
- Re-use server resource for each integration test run
### Background
Input from the input pin is consumed only once, and the default input can always be used. So when you have an input pin overriding the default input, the value will be used only once and the following run will always fall back to the default input. This can mislead the user.
Expected behaviour: the node should NOT RUN, making connected pins only use their connection(s) for sources of data.
### Changes 🏗️
* Make pin connection the mandatory source of input and not falling back to default value.
* Fix the type flakiness on block input & output. Unify the typing for BlockInput & BlockOutput using the right alias to avoid wrong typing.
* Add comment on alias
* automated test on the new behaviour.
### Background
Add formatter & linter command.
Tools: ruff --> isort --> black --> pyright.
### Changes 🏗️
Introduced:
* `poetry run format`
* `poetry run lint`
`poetry run lint` will be executed on CI.
### Background
The main scope of this change is enhancing the system capability (by fixing bug, correcting execution behaviour) to allow for creating a graph with a loop, to allow the use case of block auto-generation agent.
### Changes 🏗️
* Main changes: Add block_autogen.py (block auto-generation agent graph example).
* Refactor on test boilerplate: introduced `util/test` for initiating a server, and waiting graph execution.
* Removing unnecessary db lookup and duplicated codes used for sending execution updates on agent executor.
* Removed redundant code on test and cli code.
* Moved block test execution helper into the main code (so blockinstallerblock can use it).
* Eliminate the need of explicitly add a module into the `AVAILABLE_BLOCKS` list, any block class under the `block` folder will be auto-discovered.
* Few patches on the existing blocks.
1. Add graph versioning functionality:
- Add `version`, `isActive` fields in the `AgentGraph` model
- Add `agentGraphVersion` field in related models
- Amend & add API endpoints for graph versioning (see below)
- Amend & add data layer functions (`autogpt_server.data`) to support new operations & data schema
2. Add graph template functionality:
- Add `isTemplate` fields in the `AgentGraph` model
- Add `GraphMeta` model for template/graph metadata
- Add API endpoints for template management (see below)
- Amend & add data layer functions (`autogpt_server.data`) to support new operations & data schema
3. Enhance graph creation:
- Amended `create_graph` route to handle template-based graph creation
4. Code refactoring:
- Improved import statements
- Enhanced error handling in graph creation
5. Minor improvements:
- Add validator to auto-assign `Graph.id` if not set
## API Changes
New endpoints:
- `GET /templates`: Retrieve all templates (metadata only)
- `POST /templates`: Create a new template
- `PUT /graphs/{graph_id}`: Create a new version of a graph
- `GET /templates/{graph_id}`: Get a specific template
- `PUT /templates/{graph_id}`: Create a new version of a graph template
- `GET /graphs/{graph_id}/versions`: Get all versions of a graph
- `GET /templates/{graph_id}/versions`: Get all versions of a graph template
- `GET /graphs/{graph_id}/versions/{version}`: Get a specific graph version
- `PUT /graphs/{graph_id}/versions/active`: Set active graph version
Modified endpoints:
- `POST /graphs`: Now supports creating graphs directly from templates
- `GET /graphs/{graph_id}`: Add `version` query parameter
- `GET /graphs/{graph_id}/executions`: Add `graph_version` query parameters
## UI changes
- Improve `/build` / `FlowEditor` save mechanism
- Implement updating current agent instead of creating a new agent on every save
- Add check to only save a new version if local graph has been edited
- Add `deepEquals` function to lib/utils.ts
- Add version indicators and selector on `/monitor`
- Add shad/cn `DropdownMenu` component
- Update `AutoGPTServerAPI` client
- Update input/output types with added attributes (see above)
- Add parameter `version` to `getFlow`
- Add parameter `flowVersion?` to `listFlowRunIDs`
- Add endpoint `updateFlow(flowID, FlowUpdateable)`
- Add endpoint `createFlow(fromTemplateID, templateVersion)` (overload)
- Add endpoint `getFlowAllVersions(id)`
- Add endpoint `setFlowActiveVersion(flowID, version)`
* Refactor on the link structure and API
* Refactor on the link structure and API
* Cleanup IDS
* Remove run_id
* Update block interface
* Added websockets dependency
* Adding routes
* Adding in websocket code
* Added cli to test the websocket
* Added an outline of the message formats I plan on using
* Added webscoket message types
* Updated poetry lock
* Adding subscription logic
* Updating subscription mechanisms
* update cli
* Send updates to server
* Get single execution data
* Fix type hints and renamed function
* add callback function and type hints
* fix type hints
* Updated manager to use property
* Added in websocket updates
* Added connection manager tests
* Added tests for ws_api
* trying to work around process issues
* test formatting
* Added a create and execute command for the cli
* Updated send format
* websockets command working
* cli update
* Added model.py
* feat: Update server.py and manager.py
- Initialize blocks in AgentServer lifespan context
- Remove unnecessary await in AgentServer get_graph_blocks
- Fix type hinting in manager.py
- Validate input data in validate_exec function
* fix tests
* feat: Add autogpt_server.blocks.sample and autogpt_server.blocks.text modules
This commit adds the `autogpt_server.blocks.sample` and `autogpt_server.blocks.text` modules to the project. These modules contain blocks that are used in the execution of the Autogpt server. The `ParrotBlock` and `PrintingBlock` classes are imported from `autogpt_server.blocks.sample`, while the `TextFormatterBlock` class is imported from `autogpt_server.blocks.text`. This addition enhances the functionality of the server by providing additional blocks for text processing and sample operations.
* fixed circular import issue
* Update readme
---------
Co-authored-by: Zamil Majdy <zamil.majdy@agpt.co>
* feat(autogpt_builder): Add `AutoGPTServerAPI` client
* migrate API calls in Flow.tsx to new API client
* feat(autogpt_server): Add `/graphs/{graph_id}/executions` endpoint
In `data/execution.py`:
- Add `list_executions` function
- Rename `get_executions` to `get_execution_results`
In `server/server.py`:
- Add route
- Add `AgentServer.list_graph_runs`
- Rename `AgentServer.get_executions` to `get_run_execution_results`
* feat(autogpt_builder): Add `listFlowRunIDs` endpoint to `AutoGPTServerAPI` client
* Move `Schema` to `types.ts` and rename to `ObjectSchema`
* feat(rnd): Add type hint and strong pydantic type validation for block input/output + add reddit agent-blocks.
* feat(rnd): Add type hint and strong pydantic type validation for block input/output + add reddit agent-blocks.
* Fix reddit block
* Fix serialization
* Eliminate deprecated class property
* Remove RedditCredentialsBlock
* Cache jsonschema computation, add dictionary construction
* Add dict_split and list_split to output, add more blocks
* Add objc_split for completeness, int both input and output
* Update reddit block
* Add reddit test (untested)
* Resolved json issue on pydantic
* Add creds check on client
* Add dict <--> pydantic object flexibility
* Fix error retry
* Skip reddit test
* Code cleanup
* Chang prompt
* Make this work
* Fix linting
* Hide input_links and output_links from Node
* Add docs
---------
Co-authored-by: Aarushi <50577581+aarushik93@users.noreply.github.com>
On AgentServer, To create a Block like StringFormatterBlock or LllmCallBlock, we need some way to dynamically link input pins and aggregate them into a single list input. This will give a better experience for the user to construct an input and link it from the output of the other nodes. The scope of this change is adding support for that in the least intrusive way.
Proposal
To differentiate the input list name and its singular entry we are using the $_<index> prefix. For example:
For the input items: list[int], you can set a pin items with values like [1,2,3,4]. But you can also add input pins like items_$_0 or items_$_4 with values 1 or 2, which will be appended to the items input in alphabetical order.
The execution engine will guarantee to wait for the execution until all the input pin value is produced, so input pin with list input will produce fix-sized list.
### Background
The current implementation of AgentServer doesn't allow for a single pin to be connected to multiple nodes, this will be problematic when you have a single output node that needs to be propagated into many nodes. Or multiple nodes that possibly feed the data into a single pin (first come first serve).
This infra change is also part of the preparation for changing the `block` interface to return a stream of output instead of a single output. Treating blocks as streams requires this capability.
### Changes 🏗️
* Update block run interface from returning `(output_name, output_data)` to `Generator[(output_name, output_data)]`
* Removed `agent` term in the API, replace it with `graph` for consistency.
* Reintroduced `AgentNodeExecutionInputOutput`. `AgentNodeExecution` input & output will be a list of `AgentNodeExecutionInputOutput` which describes the input & output data of its execution. Making an execution has 1-many relation to its input output data.
* Propagating the relation and block interface change into the execution engine.
### Background
Agent execution should be able to be triggered in a recurring manner.
This PR introduced an ExecutionScheduling service, a process responsible for managing the execution schedule and triggering its execution based on a predefined cron expression.
### Changes 🏗️
* Added `scheduler.py` / `ExecutionScheduler` implementation.
* Added scheduler test.
* Added `AgentExecutionSchedule` table and its logical model & prisma queries.
* Moved `add_execution` from API server to `execution_manager`
### Background
This PR adds support on IPC on autogpt_server.
To make this happen, there are a couple of refactoring efforts being made (will be described in the `Changes` section).
Currently, there are three independent processes:
```
AgentServer ----> ExecutionManager
|
--> ExecutionScheduler
```
### Changes 🏗️
* Added Pyro5 for IPC support.
* Introduced `AppService`: a class to construct an independent process that can expose a method to other running processes (this is analogous to a microservice).
* Introduced `AppProcess`: used by `AppService` a class for creating a child process that can be executed in the background.
* Adapting existing codebase to user `AppService`.
