Ricochet

The Hybrid AI Coding Agent
Pair-program in VS Code or control your environment remotely via messengers.

Ricochet is an open-source autonomous agent designed for complex coding tasks. Unlike standard autocomplete tools, it manages its own context, learns from project patterns, and plans actions using a DAG-based architecture.

🚀 Capabilities

1. Autonomous Agent

Ricochet uses a powerful local core (written in Go) to orchestrate complex coding tasks.

• Swarm Mode (Parallel Execution): Uses a DAG-based planner to spawn multiple workers (up to 5+) for handling independent tasks simultaneously.
• Plan Mode (Task Workspaces): A dedicated planning engine that tracks task lifecycle (pending, active, verification) and persists plans (PLAN.md, CONTEXT.md) across sessions to prevent agent amnesia.
• Reflex Engine (4-Level Context): Automatically manages context windows, condensing conversation history to maintain long-term memory during deep coding sessions.
• Shadow Git: Every task has a hidden git checkpoint. Instantly undo/redo AI-generated code without polluting your main project history.
• Skill Injector: Detects your current task (e.g., "working on backend controllers") and automatically injects relevant project guidelines and best practices.
• Auto-QC: Automatically runs build and lint checks after editing code. If the build fails, Ricochet catches the error and attempts to fix it before returning control to you.

2. Live Mode (Ether)

Don't be tied to your desk. Toggle "Live Mode" to connect Ricochet to a Telegram or Discord bot.

• Remote Control: Ask your agent to "fix the bug" or "deploy to staging" while you are away.
• Notifications: Receive real-time updates when tasks are completed or if the agent needs clarification.
• Voice Support: Send voice messages to your agent for natural language prompting.

3. Tooling & Integration

• CLI & VS Code: Full feature parity between the Visual Studio Code extension and the standalone Terminal User Interface (TUI).
• MCP Support: Fully compatible with the Model Context Protocol. Connect any MCP server (GitHub, Postgres, Filesystem) to extend Ricochet's capabilities.
• Cross-Platform: Runs natively on macOS, Linux, and Windows.
• Multi-Provider: Bring Your Own Key (BYOK). Supports Anthropic (Claude), OpenAI (GPT-4), Google (Gemini), DeepSeek, and OpenRouter.

📦 Installation

VS Code Extension (Recommended)

Install Ricochet directly from the VS Code Marketplace.

CLI Tool

The standalone CLI is bundled with the extension (or can be built from source). To launch the terminal interface:

# Launch Ricochet CLI
ricochet

🔑 Setup (DeepSeek)

Currently, Ricochet interaction is handled through DeepSeek.

1. Get an API Key: Obtain your key from the DeepSeek Platform.
2. Configure:
   • VS Code: Enter the key in the extension settings.
   • CLI: Run ricochet /init or set the DEEPSEEK_API_KEY environment variable.

Note: BYOK (Bring Your Own Key) support for other providers (Anthropic, OpenAI, Gemini) is in progress.

🛠 Support the Project

Ricochet is developed solo and is completely open-source. If it helps you build faster, please consider supporting its development.

• Star the Repo: github.com/Grik-ai/ricochet
• Ko-fi: ko-fi.com/igoryan34
• PayPal: Donate via PayPal
• Crypto:
  • TON: UQB93GTsF6ZI7ljBViLr-IHIf93HpqwolC51jR5Und7GAwm4
  • USDT (TRC20): TRykhiHNeXxcdmhn5DBP2GGLnKQDBVzjhB
  • USDT (SOL): DAaY5J6tg6PY7M77mxpMxGFMqLsCtPRKBmgmmCS9zM5b
  • USDT (ERC20): 0x5eB4C13497e10849724714D351C139Fc6Ab00Adc

Star History

License

Apache 2.0 © 2025 Igor Pryimak