Developer Releases Self-Hosted AI Assistant Built Entirely in Rust
Fabien Penso, a principal engineer with 25 years of production systems experience, released Moltis on February 12, 2026. The AI assistant is delivered as a single Rust binary measuring approximately 60MB with 150,000 lines of code, requiring no Node, Python, or runtime dependencies. The project received 65 points and 23 comments on Hacker News, while the announcement on X garnered 592 likes, 78 retweets, and 761 bookmarks.
Architecture Prioritizes Privacy and Self-Hosting
Moltis distinguishes itself through its minimal dependency footprint and privacy-first design. The assistant includes no telemetry and sends no data to external servers. Penso applies his philosophy of owning your content and infrastructure to AI systems, stating: "If something touches your files, credentials, and daily workflow, you should be able to inspect it, audit it, and fork it if the project changes direction."
The system supports multi-provider LLM routing including OpenAI, local GGUF and MLX models, and Hugging Face. Code execution runs in sandboxed Docker, Podman, or Apple Containers for safety. Deployment options include 1-click deploys on DigitalOcean and Fly.io, with Penso demonstrating 50-second deployment times.
Self-Extending Skills Enable Runtime Capability Growth
Moltis implements self-extending skills inspired by Pi, allowing the agent to create its own capabilities at runtime. The system maintains hybrid vector and full-text memory that persists across sessions, with MCP tool servers featuring auto-restart capabilities for reliable integration. Users can access Moltis through a web UI, Telegram bot, or API, all sharing the same context.
The project includes built-in OpenTelemetry and Prometheus support for full observability. Penso describes the current state as "alpha," noting he uses it daily and released it because "it's useful, not because it's done." The MIT-licensed project is fully open source.
Developer Background Reflects Broader AI Agent Movement
Penso's experience spans Ruby and Swift before moving to Rust. He has documented his observations of the AI agent community, noting coworking spaces hosting group installations of autonomous AI systems reminiscent of "2000s Linux install parties." He characterized the current AI shift as "generational, bigger than the birth of the Internet," emphasizing the need to "adapt or get left behind."
Key Takeaways
- Moltis is a single 60MB Rust binary with 150,000 lines of code requiring no Node, Python, or runtime dependencies
- The system supports multiple LLM providers including OpenAI, local GGUF/MLX models, and Hugging Face with sandboxed code execution
- Self-extending skills allow the agent to create its own capabilities at runtime, with persistent memory across sessions
- The MIT-licensed project includes no telemetry, 1-click cloud deploys, and full observability through OpenTelemetry and Prometheus
- Developer Fabien Penso has 25 years of production systems experience and built Moltis to apply his philosophy of owning critical infrastructure