Moltis
A self-hosted AI assistant built in Rust by Fabien Penso. It prioritizes observability and security architecture over plugin breadth. Where OpenClaw runs everything in one Node.js process, Moltis uses 27 workspace crates with explicit trait definitions for every integration boundary.
Key Facts
- Language: Rust
- Footprint: 44MB self-contained binary; web UI and assets included
- Architecture: Gateway-centric; crates for gateway, agents, tools, memory, MCP, voice, sessions, projects, plugins
- Channels: Web UI, Telegram, API
- Tooling: MCP over stdio and HTTP/SSE; hook lifecycle; parallel tool execution; container-based sandbox (Docker, Podman, Apple Container)
- Observability: Prometheus metrics, OpenTelemetry tracing (OTLP), structured logging
- Security: Password + passkey (WebAuthn), API key auth, endpoint throttling, WS origin checks, TLS
- Other: Tailscale integration, local LLM support, voice I/O (15+ providers), hybrid memory search
- License: MIT
Best For
Individual developers and small technical teams who want a self-hosted agent they can trust in production, with serious observability tooling and no architectural shortcuts. When something breaks, Moltis gives you clear signals; OpenClaw does not.
Note: Sub-agent delegation is on the roadmap but not yet merged. The community is smaller than OpenClaw's. Documentation is still catching up. No RBAC, audit trail, or enterprise compliance certifications.
Compared to OpenClaw
Moltis behaves more like a local AI gateway platform than a simple bot. Its multi-crate structure supports modular ownership and long-horizon evolution. If you need production-grade observability and a deliberately secure architecture in a self-hosted agent, Moltis is the strongest fit among the open-source options.