adolfousier/opencrabs
The all-in-one AI agent living in your terminal. Build landing, mobile apps, backends, manages files, deep research, schedule tasks and events. Self-improving, self-healing, fully autonomous. TUI, CLI and Daemon. Connect to your favorite channels.
OpenCrabs – a self‑contained, Rust‑based AI agent
What it is – OpenCrabs is a terminal‑only, single‑binary AI assistant written in Rust. It talks directly to LLM providers (OpenAI, Anthropic, Gemini, Claude, Ollama, etc.) over HTTPS, stores all data locally (SQLite chats, embeddings, brain files, API keys) and never runs any background server. The project bundles a TUI built on ratatui and optional Docker images for easy deployment.
Key capabilities
| Category | Highlights |
|---|---|
| Multi‑provider support | Works with dozens of commercial and local models; you can chain fallback providers and switch models per session. |
| Zero‑telemetry, local‑first | All chats, tool executions, and memory stay on the host. Outbound traffic is limited to the explicit LLM/API calls you trigger. |
| 3‑tier memory system | • Brain files (user‑curated markdown) • Daily log files with auto‑summaries • Hybrid search (FTS5 + vector embeddings) that can run offline with a tiny local embedding model or use any remote /v1/embeddings endpoint. |
| Dynamic tool system | Built‑in tools for bash, git, web search, GitHub CLI, browser automation (Chromium CDP), and a generic HTTP tool defined in tools.toml. |
| Multi‑agent orchestration | Spawn child agents (explore, code, research, etc.) that run in isolated sessions and can be coordinated via spawn_agent, wait_agent, … commands. |
| Self‑improvement loop (experimental) | Records tool usage, analyses feedback, and can automatically edit its own brain files, logging every change. |
| Multimodal input | Images (paths, URLs, clipboard) and videos can be attached; the agent calls provider‑specific vision tools (e.g., Gemini) to analyse them. |
| Local speech | Optional offline Whisper‑cpp transcription and TTS, all bundled in ~130 MB. |
| Usage dashboard | /usage shows token counts, cost estimates, and activity charts per project/provider. |
| Security‑first design | API keys are zeroised from RAM, no inbound listeners, and the binary is a single compiled artifact (~35 MB). |
How to get started
- Install the binary – either download a pre‑built release, build with
cargo install opencrabs, or run the provided Docker image. - Run
opencrabs– the TUI launches with an onboarding wizard that walks you through creatingkeys.toml(your provider API keys) and optionalconfig.toml. - Start a chat. The system prompt is assembled from the brain markdown files (
SOUL.md,AGENTS.md,TOOLS.md, …) which you can edit at any time. - Use slash commands (
/goal,/cron,/tool …) to trigger built‑in features, or attach files for multimodal processing.
Documentation & community – Full docs are hosted at https://docs.opencrabs.com (six languages) and the project’s GitHub page includes benchmarks, architecture diagrams, and a “Full onboard” video. The repo is actively maintained (CI badge, Trendshift rankings) and welcomes contributions.
OpenCrabs is a genuine, open‑source AI‑agent framework focused on privacy, performance, and extensibility, built entirely in Rust.
Related
- Project
- Project
- Project
- Project
- Project