Skip to content

CHANGELOG.md

Latest commit

 

History

History
219 lines (178 loc) · 6.62 KB

CHANGELOG.md

File metadata and controls

219 lines (178 loc) · 6.62 KB
Raw

Changelog

All notable changes to AutoCodeAI will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[2.0.0] - 2024-04-08

🎉 Major Release - Parallel Execution, Tool Integration, and Flexible LLM Support

Added

Tool Integration

  • Tool Executor System (core/tools/tool_executor.py)
    • Whitelist-based command execution for security
    • Support for git operations (clone, status, add, commit, push, pull, diff, log)
    • Package management (pip install, uninstall, list, freeze)
    • File operations (ls, mkdir, cat, rm, cp, mv)
    • Python execution (run scripts, pytest)
    • Environment utilities (env, pwd, which)
    • Parameter sanitization with shlex.quote
    • Timeout enforcement and error handling
    • Working directory isolation

Parallel Execution

  • Parallel Agent Execution in orchestrator
    • Independent tasks run concurrently
    • Automatic dependency management via parallel_group in plan steps
    • 3-5x faster completion for multi-step tasks
    • run_parallel() method for explicit parallel execution
    • run_parallel_coders() helper for multiple coder agents
  • New API Endpoint: POST /api/agent/run_parallel
    • Execute multiple steps concurrently
    • Supports mixed agent types (coder, tester, tool)
    • Returns results array preserving order

Flexible LLM Support

  • Multi-Mode LLM Client (core/utils/llm.py)
    • LiteLLM mode (default): unified interface to 100+ providers
    • OpenAI mode: direct OpenAI API access
    • DeepSeek mode: direct DeepSeek API integration
    • Local mode: Ollama and compatible local endpoints
    • Per-agent model routing maintained across all modes
    • Seamless mode switching via environment variables
  • Configuration Variables:
    • LLM_MODE: Choose provider mode
    • OPENAI_API_KEY, DEEPSEEK_API_KEY: Provider-specific keys
    • LOCAL_LLM_URL, LOCAL_MODEL: Local model configuration
    • LLM_MODEL: Global model override

Web User Interface

  • Modern Web UI (static/index.html)
    • Beautiful, responsive design with dark theme
    • Real-time output streaming via SSE
    • Task submission with file context
    • Status indicators (idle, running, success, error)
    • Clear output functionality
    • Keyboard shortcuts (Shift+Enter to submit)
    • Feature showcase cards
  • Static File Serving in main.py
    • Root redirect to web UI
    • FastAPI StaticFiles mount

Enhanced Planning

  • Updated PlannerAgent
    • Support for tool agent type in plans
    • parallel_group field for concurrent execution
    • tool_name and tool_params for tool steps
    • Improved JSON schema documentation

Changed

  • Orchestrator (services/orchestrator.py)

    • Refactored run() method to support parallel execution
    • Added _execute_step() helper for unified step execution
    • Integrated ToolExecutor for tool steps
    • Groups steps by parallel_group for concurrent execution
    • Enhanced error handling for parallel tasks
    • Tool output stored in memory for downstream agents

  • API Routes (api/routes.py)

    • Added ParallelRequest model
    • Enhanced documentation with examples
    • Updated route descriptions

  • Main Application (main.py)

    • Updated title to "AutoCodeAI"
    • Version bumped to 2.0.0
    • Added static file serving
    • Root endpoint redirects to web UI
    • Enhanced health endpoint with version info

  • Requirements (requirements.txt)

    • Added explicit openai>=1.30.0 dependency
    • Updated LiteLLM dependency documentation

  • Environment Configuration (.env.example)

    • Added LLM_MODE configuration section
    • Added tool executor settings
    • Added parallel execution settings
    • Reorganized for better clarity
    • Added inline documentation

Improved

  • Documentation (README.md)

    • Added "What's New in v2.0" section
    • Updated project name to AutoCodeAI
    • Enhanced architecture description
    • Added tool executor to agents table
    • Updated project structure with new files
    • Comprehensive configuration table with sections
    • Web UI usage examples
    • Parallel execution examples
    • LLM mode switching examples
    • Updated API reference with new endpoints

  • Error Handling

    • Better error messages in LLM client
    • Tool executor validation and sanitization
    • Parallel execution exception handling

  • Performance

    • Concurrent task execution for independent steps
    • Reduced latency for multi-step workflows

Security

  • Tool Executor
    • Whitelist-only command execution
    • Parameter sanitization prevents shell injection
    • Timeout enforcement prevents runaway processes
    • Working directory isolation

Dependencies

  • openai>=1.30.0 - Direct API access for non-LiteLLM modes
  • All existing dependencies maintained

[1.0.0] - 2024-03-XX

Initial Release

  • Multi-agent orchestration (Planner, Coder, Tester, Debugger, Critic)
  • Docker sandbox execution
  • ChromaDB semantic memory
  • Real-time streaming (SSE, WebSocket)
  • Diff-based file editing
  • Repository awareness and indexing
  • LiteLLM integration for multiple providers
  • REST API endpoints
  • Comprehensive test suite

Upgrade Guide: 1.0 → 2.0

Breaking Changes

None! Version 2.0 is fully backward compatible.

Recommended Actions

  1. Update environment configuration:

    cp .env.example .env
# Add new variables: LLM_MODE, TOOL_EXECUTOR_TIMEOUT, etc.

  2. Try the web UI:

    uvicorn main:app --reload
open http://localhost:8000

  3. Test parallel execution:

    # Plans can now include parallel_group
{
  "steps": [
    {"agent": "coder", "description": "Task 1", "parallel_group": 1},
    {"agent": "coder", "description": "Task 2", "parallel_group": 1},
    {"agent": "tester", "description": "Test", "parallel_group": 2}
  ]
}

  4. Try tool integration:

    # Plans can now include tool steps
{
  "steps": [
    {"agent": "tool", "tool_name": "git_clone", 
     "tool_params": {"url": "...", "dest": "lib"}},
    {"agent": "coder", "description": "Use the library"}
  ]
}

  5. Experiment with different LLM modes:

    # Try DeepSeek
export LLM_MODE=deepseek
export DEEPSEEK_API_KEY=sk-...

# Or local models
export LLM_MODE=local
export LOCAL_MODEL=deepseek-coder

Migration Notes

  • All existing API endpoints remain unchanged
  • Existing plans without parallel_group execute sequentially (backward compatible)
  • Default LLM_MODE is litellm (existing behavior)
  • Tool execution is opt-in via plan steps