mirror of
https://github.com/Significant-Gravitas/AutoGPT.git
synced 2026-02-10 23:05:17 -05:00
d33459ddb5
## Summary
This PR introduces a complete cloud storage infrastructure and file
upload system that agents can use instead of passing base64 data
directly in inputs, while maintaining backward compatibility for the
builder's node inputs.
### Problem Statement
Currently, when agents need to process files, they pass base64-encoded
data directly in the input, which has several limitations:
1. **Size limitations**: Base64 encoding increases file size by ~33%,
making large files impractical
2. **Memory usage**: Large base64 strings consume significant memory
during processing
3. **Network overhead**: Base64 data is sent repeatedly in API requests
4. **Performance impact**: Encoding/decoding base64 adds processing
overhead
### Solution
This PR introduces a complete cloud storage infrastructure and new file
upload workflow:
1. **New cloud storage system**: Complete `CloudStorageHandler` with
async GCS operations
2. **New upload endpoint**: Agents upload files via `/files/upload` and
receive a `file_uri`
3. **GCS storage**: Files are stored in Google Cloud Storage with
user-scoped paths
4. **URI references**: Agents pass the `file_uri` instead of base64 data
5. **Block processing**: File blocks can retrieve actual file content
using the URI
### Changes Made
#### New Files Introduced:
- **`backend/util/cloud_storage.py`** - Complete cloud storage
infrastructure (545 lines)
- **`backend/util/cloud_storage_test.py`** - Comprehensive test suite
(471 lines)
#### Backend Changes:
- **New cloud storage infrastructure** in
`backend/util/cloud_storage.py`:
- Complete `CloudStorageHandler` class with async GCS operations
- Support for multiple cloud providers (GCS implemented, S3/Azure
prepared)
- User-scoped and execution-scoped file storage with proper
authorization
- Automatic file expiration with metadata-based cleanup
- Path traversal protection and comprehensive security validation
- Async file operations with proper error handling and logging
- **New `UploadFileResponse` model** in `backend/server/model.py`:
- Returns `file_uri` (GCS path like
`gcs://bucket/users/{user_id}/file.txt`)
- Includes `file_name`, `size`, `content_type`, `expires_in_hours`
- Proper Pydantic schema instead of dictionary response
- **New `upload_file` endpoint** in `backend/server/routers/v1.py`:
- Complete new endpoint for file upload with cloud storage integration
- Returns GCS path URI directly as `file_uri`
- Supports user-scoped file storage for proper isolation
- Maintains fallback to base64 data URI when GCS not configured
- File size validation, virus scanning, and comprehensive error handling
#### Frontend Changes:
- **Updated API client** in
`frontend/src/lib/autogpt-server-api/client.ts`:
- Modified return type to expect `file_uri` instead of `signed_url`
- Supports the new upload workflow
- **Enhanced file input component** in
`frontend/src/components/type-based-input.tsx`:
- **Builder nodes**: Still use base64 for immediate data retention
without expiration
- **Agent inputs**: Use the new upload endpoint and pass `file_uri`
references
- Maintains backward compatibility for existing workflows
#### Test Updates:
- **New comprehensive test suite** in
`backend/util/cloud_storage_test.py`:
- 27 test cases covering all cloud storage functionality
- Tests for file storage, retrieval, authorization, and cleanup
- Tests for path validation, security, and error handling
- Coverage for user-scoped, execution-scoped, and system storage
- **New upload endpoint tests** in `backend/server/routers/v1_test.py`:
- Tests for GCS path URI format (`gcs://bucket/path`)
- Tests for base64 fallback when GCS not configured
- Validates file upload, virus scanning, and size limits
- Tests user-scoped file storage and access control
### Benefits
1. **New Infrastructure**: Complete cloud storage system with
enterprise-grade features
2. **Scalability**: Supports larger files without base64 size penalties
3. **Performance**: Reduces memory usage and network overhead with async
operations
4. **Security**: User-scoped file storage with comprehensive access
control and path validation
5. **Flexibility**: Maintains base64 support for builder nodes while
providing URI-based approach for agents
6. **Extensibility**: Designed for multiple cloud providers (GCS, S3,
Azure)
7. **Reliability**: Automatic file expiration, cleanup, and robust error
handling
8. **Backward compatibility**: Existing builder workflows continue to
work unchanged
### Usage
**For Agent Inputs:**
```typescript
// 1. Upload file
const response = await api.uploadFile(file);
// 2. Pass file_uri to agent
const agentInput = { file_input: response.file_uri };
```
**For Builder Nodes (unchanged):**
```typescript
// Still uses base64 for immediate data retention
const nodeInput = { file_input: "data:image/jpeg;base64,..." };
```
### Checklist 📋
#### For code changes:
- [x] I have clearly listed my changes in the PR description
- [x] I have made a test plan
- [x] I have tested my changes according to the test plan:
- [x] All new cloud storage tests pass (27/27)
- [x] All upload file tests pass (7/7)
- [x] Full v1 router test suite passes (21/21)
- [x] All server tests pass (126/126)
- [x] Backend formatting and linting pass
- [x] Frontend TypeScript compilation succeeds
- [x] Verified GCS path URI format (`gcs://bucket/path`)
- [x] Tested fallback to base64 data URI when GCS not configured
- [x] Confirmed file upload functionality works in UI
- [x] Validated response schema matches Pydantic model
- [x] Tested agent workflow with file_uri references
- [x] Verified builder nodes still work with base64 data
- [x] Tested user-scoped file access control
- [x] Verified file expiration and cleanup functionality
- [x] Tested security validation and path traversal protection
#### For configuration changes:
- [x] No new configuration changes required
- [x] `.env.example` remains compatible
- [x] `docker-compose.yml` remains compatible
- [x] Uses existing GCS configuration from media storage
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
---------
Co-authored-by: Claude AI <claude@anthropic.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Nicholas Tindle <nicholas.tindle@agpt.co>