Use Claude Code / CodeX CLI to perform multiple tasks in parallel with a Codex-style UI.

A code agent task management system that provides parallel execution of AI-powered coding tasks. Users can run multiple Claude Code agents simultaneously through a Codex-style web interface, with support for different agents for comparison and evaluation.

• 🤖 Multi-Agent Support: Run Claude Code and other AI agents in parallel
• 🔄 Parallel Task Management: Execute multiple coding tasks simultaneously
• 🌐 Codex-Style Web UI: Clean interface for managing agent tasks
• 🔍 Agent Comparison: Compare outputs from different AI models
• 🐳 Containerized Execution: Secure sandboxed environment for each task
• 🔗 Git Integration: Automatic repository cloning, commits, and PR creation
• Selfhost: Deploy you rown parallel code agent platform.

• Frontend: Next.js with TypeScript and TailwindCSS
• Backend: Python Flask API with Docker orchestration
• Agents: Claude Code (Anthropic) with extensible support for other models
• Task Management: Parallel execution system based on container

The system implements comprehensive security measures to ensure safe execution of AI agents:

• Non-root Execution: All containers run as UID 1000 (non-root user)
• No Privileged Access: Containers run without --privileged flag or host PID namespace
• Capability Restrictions: No dangerous Linux capabilities granted (no CAP_SYS_ADMIN, etc.)
• Privilege Escalation Prevention: no-new-privileges=true security option enforced
• Sandboxed Environment: Proper namespace isolation between container and host

• Volume Permissions: Workspace volumes mounted with proper ownership (1000:1000)
• Read-only System Paths: Containers cannot modify /etc, /usr, or other system directories
• Isolated Workspaces: Each task gets its own temporary workspace with restricted permissions

The project includes security test scripts:

• test-container-security.sh: Verifies container isolation and restrictions
• test-ai-functionality.sh: Ensures agents work correctly with security measures
• Unit tests in server/tests/test_container_security.py

1. Setup

   git clone <this-repo>
   cd async-code
   ./build.sh

2. Configure

   • Add your Anthropic API key to server/.env
   • Get a GitHub Personal Access Token with repo permissions

3. Run

   • Frontend: http://localhost:3000
   • Backend API: http://localhost:5000

1. Setup GitHub Token: Enter your GitHub token in the web interface
2. Configure Repository: Specify target repository and branch
3. Select Agent: Choose your preferred AI agent (Claude Code, etc.)
4. Submit Tasks: Start multiple coding tasks in parallel
5. Compare Results: Review and compare outputs from different agents
6. Create PRs: Generate pull requests from successful tasks

See Environment Variables Documentation for complete configuration details.

# server/.env (Required)
SUPABASE_URL=your_supabase_url
SUPABASE_SERVICE_ROLE_KEY=your_service_role_key
JWT_SECRET=your-very-secure-secret-key-minimum-32-chars

# API Keys (Optional)
ANTHROPIC_API_KEY=your_anthropic_api_key_here

# Run all services
docker-compose up

# Development mode
cd async-code-web && npm run dev  # Frontend
cd server && python main.py      # Backend

The project includes a comprehensive test user system for automated testing:

• Test User Management: Automated test user creation with JWT authentication
• Local Git Repositories: Tests use local repos in /tmp/test-repos/ to avoid external dependencies
• Isolated Test Environment: Complete separation from production data
• Automated Cleanup: Test data automatically cleaned up after 1 hour

# Run tests with test profile
docker-compose -f docker-compose.yml -f docker-compose.test.yml up

# Run Playwright E2E tests
cd async-code-web && npm run test:e2e

# Run backend unit tests
cd server && pytest

See LICENSE file for details.