A local autonomous software engineering agent that works in the background.

Cirrus is an enhanced SWE (Software Engineering) agent that can:

• Clone repositories and create isolated working environments
• Generate intelligent implementation plans using vector embeddings
• Execute iterative implementation with verification
• Create pull requests from feature branches
• Clean up temporary resources automatically

The agent follows a structured 10-phase workflow:

graph TD
    A[TASK] --> B(Clone repository)
    B --> C(Vector embeddings)
    C --> D["Plan (from task and using the embeddings)"]
    D --> E["Iterative Implementation (shift to new branch and commit)"]
    E --> F(Task Verification)

    F -- Does not verify --> E
    F -- Verifies --> G{Error Fixing}

    G -- If errors --> E
    G -- No errors --> H(Create a PR from new branch)

    H --> I(Delete temp files)
    I --> J[Task done]

• Uses vector embeddings to understand codebase context
• Generates detailed implementation plans with specific file modifications
• Considers existing code patterns and architecture

• Creates feature branches automatically
• Makes targeted file modifications based on the plan
• Iterates with verification until task completion

• Built-in verification and testing phases
• Error detection and automatic fixing
• Comprehensive logging for debugging

• Full git workflow support (clone, branch, commit, push, PR)
• Temporary directory management
• Clean resource cleanup

from cirrus_agent.agents.enhanced_swe_agent import EnhancedSWEAgent

# Initialize the agent
agent = EnhancedSWEAgent(
    model_config={
        "api_key": "your-openai-api-key",
        "base_url": "https://api.openai.com/v1"  # optional
    },
    github_token="your-github-token"
)

# Execute a task
result = await agent.execute_task(
    task_description="Update Node.js to v22 and upgrade Next.js",
    repo_url="https://github.com/username/repository"
)

• Dependency Updates: "Update React to v19 and Next.js to latest"
• Configuration Changes: "Add dark mode support with Tailwind CSS"
• Code Refactoring: "Refactor authentication middleware to use JWT tokens"
• Feature Implementation: "Add user profile page with avatar upload"

• EnhancedSWEAgent: Main orchestrator that manages the workflow
• RepositoryManager: Handles repository cloning and cleanup
• EmbeddingService: Creates and manages code embeddings for context
• ValidationEngine: Verifies implementation and detects issues
• AutonomousToolSystem: Orchestrates autonomous tool execution

• File Operations: read, write, update, search files
• Git Operations: branch, commit, push, status, create PR
• Code Analysis: vector embeddings, semantic search
• Validation: testing, linting, type checking

# Required
OPENAI_API_KEY=your-openai-api-key
GITHUB_TOKEN=your-github-token

# Optional
EMBEDDING_SERVICE_URL=http://localhost:8000
VALIDATION_ENGINE_URL=http://localhost:8001

model_config = {
    "api_key": "sk-...",
    "base_url": "https://api.openai.com/v1",  # optional
    "model": "kimi-k2-0711-preview",          # default
    "temperature": 0.1                      # for consistent results
}

# Clone the repository
git clone https://github.com/saurish/cirrus.git
cd cirrus

# Install dependencies
npm install  # for frontend
cd backend && pip install -r requirements.txt

# Backend tests
cd backend
python -m pytest tests/

# Frontend tests
cd frontend
npm test

The system follows a modular architecture with clear separation of concerns:

• Agent Layer: High-level orchestration and workflow management
• Tool Layer: Specific operations (file, git, analysis tools)
• Service Layer: Core services (embeddings, validation)
• Repository Layer: Git and file system operations

1. Fork the repository
2. Create a feature branch: git checkout -b feature-name
3. Make your changes
4. Run tests and validation
5. Submit a pull request

MIT License - see LICENSE file for details.