🦀 Personal Multi-agent AI assistant platform, written in Rust — multi-agent, self-learning skills, RAG, MCP, and agentskills.io compatible
# Add to your Claude Code skills
git clone https://github.com/opencrust-org/opencrustLast scanned: 5/30/2026
{
"issues": [],
"status": "PASSED",
"scannedAt": "2026-05-30T16:21:50.782Z",
"npmAuditRan": true,
"pipAuditRan": true
}No comments yet. Be the first to share your thoughts!
A single 16 MB binary that runs your AI agents across Telegram, Discord, Slack, WhatsApp, WhatsApp Web, LINE, WeChat, iMessage and MQTT - with encrypted credential storage, config hot-reload, and 13 MB of RAM at idle. Built in Rust for the security and reliability that AI agents demand.
# Install (Linux, macOS)
curl -fsSL https://raw.githubusercontent.com/opencrust-org/opencrust/main/install.sh | sh
# Interactive setup - pick your LLM provider and channels
opencrust init
# Start - on first message, the agent will introduce itself and learn your preferences
opencrust start
# Diagnose configuration, connectivity, and database health
opencrust doctor
# Requires Rust 1.85+
cargo build --release
./target/release/opencrust init
./target/release/opencrust start
# Optional: include WASM plugin support
cargo build --release --features plugins
Once the gateway is running, open your browser at:
http://127.0.0.1:3888
The built-in web UI lets you chat with your agent, switch LLM providers on the fly, manage MCP servers, and monitor connected channels — all without restarting.
Authentication — if
api_keyis set inconfig.yml, the UI will prompt for the gateway key before connecting.
Chat with your agent directly from the terminal — no browser needed.
Requires a running gateway. Run
opencrust init(first time only) thenopencrust startbefore usingopencrust chat.
# First-time setup
opencrust init
opencrust start # or: opencrust start -d (daemon mode)
# Open terminal chat
opencrust chat
opencrust chat --agent coder # start with a named agent
opencrust chat --url http://host:3888 # connect to a remote gateway
Chat commands: /help · /new (fresh session) · /agent <id> · /clear · /exit
Pre-compiled binaries for Linux (x86_64, aarch64), macOS (Intel, Apple Silicon), and Windows (x86_64) are available on GitHub Releases.
| | |
|---|---|
| Binary size | 16 MB single binary |
| Memory at idle | 13 MB |
| Cold start | 3 ms |
| Credential storage | AES-256-GCM encrypted vault |
| Auth default | Enabled (WebSocket pairing) |
| Scheduling | Cron, interval, one-shot |
| Multi-agent routing | Yes (named agents) |
| Session orchestration | Yes |
| MCP support | Stdio + HTTP |
| Channels | 9 |
| LLM providers | 15 |
| Pre-compiled binaries | Yes |
| Config hot-reload | Yes |
| Plugin system | WASM (sandboxed) |
| Self-update | Yes (opencrust update) |
| Security scan | ✅ skills prompt-injection scan before install |
| Self-improvement | ✅ cross-session patterns, skill lifecycle, confidence gate |
Benchmarks measured on a 1 vCPU, 1 GB RAM DigitalOcean droplet.
OpenCrust is built for the security requirements of always-on AI agents that access private data and communicate externally.
~/.opencrust/credentials/vault.json. Never plaintext on disk.--features plugins).127.0.0.1 by default, not 0.0.0.0.Native providers:
base_urlOpenAI-compatible providers:
tts-1, tts-1-hd), any OpenAI-compatible endpointauto_reply_voice: true synthesizes every text response as audio automaticallytts_max_chars limits synthesis length; long responses are truncated with a warning{"user_id","text"}, one session per device), auto-detection, exponential backoff reconnect, QoS 0/1/2, optional TLS (mqtts://)server_tool)config.yml or ~/.opencrust/mcp.json (Claude Desktop compatible)opencrust mcp list, opencrust mcp inspect <name>, opencrust mcp resources <name>, opencrust mcp prompts <name>~/.opencrust/dna.md with your name, communication style, guidelines, and the bot's own identitydna.md and the agent adapts immediatelyopencrust migrate openclaw imports your existing SOUL.md