Summary

OpenCrabs is a Rust-based, terminal-native AI orchestration layer designed for software development. It empowers developers with advanced AI capabilities directly within their terminal, supporting both cloud and local LLMs, sophisticated memory management, and seamless integration with development workflows.

How It Works

This project employs a Rust core with a Ratatui-driven TUI, offering a performant and feature-rich environment. Its architecture centers on a dynamic "brain" system, assembling agent personality and knowledge from editable Markdown files. Key innovations include a 3-tier memory system leveraging local vector embeddings for offline, private recall, and a self-sustaining capability allowing the agent to rebuild and hot-restart itself. It integrates multiple LLM providers, multimodal inputs, and messaging channels, providing a comprehensive AI assistant.

Quick Start & Requirements

OpenCrabs offers multiple installation paths: pre-built binaries for Linux, macOS, and Windows; building from source (requiring Rust nightly); or via Docker Compose. Essential prerequisites include an API key for cloud LLMs or a local LLM setup (LM Studio, Ollama). Building from source on Linux requires build-essential, pkg-config, libssl-dev, and libchafa-dev. Windows users need CMake, NASM, and Visual Studio Build Tools, or can use WSL2. Setup is streamlined by an 8-step onboarding wizard. Links to GitHub Releases and build notes are available.

Highlighted Details

• Multi-Provider LLM Support: Integrates Anthropic, OpenAI, OpenRouter (400+ models), MiniMax, and any OpenAI-compatible endpoint (Ollama, LM Studio), fetching model lists live.
• Advanced Memory System: Features a 3-tier memory architecture: user-curated MEMORY.md, auto-compacted daily logs, and a local hybrid search combining FTS5 keyword matching with 768-dim vector embeddings for offline recall.
• Self-Sustaining Agent: Capable of modifying its own source code, building, testing, and hot-restarting via Unix exec(), enabling autonomous updates.
• Multimodal & Messaging: Supports image/PDF attachments, voice I/O, and offers full Telegram, WhatsApp, and Discord bot integrations.
• Plan Mode: Enables structured task decomposition with dependency graphs, complexity ratings, and inline approval workflows for complex operations.
• Local LLM Focus: Prioritizes 100% private, zero-cost operation with local LLMs via OpenAI-compatible endpoints.

Maintenance & Community

The project is authored by Adolfo Usier. Community support is primarily channeled through GitHub Issues and Discussions. Ongoing development is indicated by upcoming features like Slack integration.

Licensing & Compatibility

OpenCrabs is released under the permissive MIT License, allowing for commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

The project is under active development and may contain bugs or incomplete features. Users are responsible for managing cloud LLM API costs. The self-updating hot-restart mechanism is Unix-only. Default binaries require AVX2; older CPUs may need custom builds or will have vector embeddings disabled. A known issue with agent hallucinating tool calls requires starting a new session, with a fix planned.