opencrabs/opencrabs

What it solves

OpenCrabs is a self-contained, autonomous AI agent that runs as a single Rust binary on your terminal. It eliminates the need for complex server infrastructure, heavy dependencies (like Node.js), and telemetry, providing a privacy-focused environment where data and API keys stay local to the user's machine.

How it works

The project is built in Rust and makes direct HTTPS calls to various LLM providers. It uses a 3-tier memory system (curated brain files, daily logs, and hybrid search combining FTS5 and vector embeddings) to maintain context. It features a "Brain System" where the agent's personality, rules, and tools are defined in editable Markdown files. For complex tasks, it can spawn specialized child agents (General, Explore, Plan, Code, Research) to execute tasks in parallel.

Who it’s for

It is designed for developers and power users who want a high-performance, private, and autonomous AI agent that integrates directly into their terminal without the overhead of a cloud-based agent framework.

Highlights

• Single Binary: No runtime or server required; zero dependencies.
• Zero Telemetry: No analytics or tracking; all data remains on the local machine.
• Multi-Provider Support: Compatible with Anthropic, OpenAI, Google Gemini, and local LLMs via Ollama or LM Studio.
• 3-Tier Memory: Combines user-curated files, automated daily logs, and hybrid vector/keyword search.
• Multi-Agent Orchestration: Ability to spawn typed child agents for parallel task execution.
• Recursive Self-Improvement: Experimental system that tracks tool execution and user corrections to autonomously update its own brain files.