moltis-org/moltis

What it solves

Moltis provides a secure, local-first way to host a persistent personal AI agent server. It eliminates the need to rely on cloud relays for your agent's data and memory, ensuring that API keys and sensitive information stay on your own hardware while providing a unified interface across multiple communication channels.

How it works

It is implemented as a single Rust binary that acts as a gateway between the user and various LLM providers (such as OpenAI Codex, GitHub Copilot, or local models). The system uses a modular architecture of 59 Rust crates to handle chat orchestration, tool execution, and session management. To ensure security, every command the agent executes is run within a sandboxed container (using Docker, Podman, Apple Container, or WASM), preventing the agent from accessing the host machine directly.

Who it’s for

Users who want a self-hosted, private AI agent that can integrate with their existing messaging apps (Telegram, Discord, Signal, etc.) and automate tasks via browser automation or MCP servers without sacrificing security or data privacy.

Highlights

• Secure Sandboxing: All tool executions are isolated in containers to prevent host-level attacks.
• Multi-Channel Access: Native integrations with Telegram, Signal, Discord, Slack, and a dedicated Web UI.
• Persistent Memory: Uses SQLite with full-text and vector search for long-term recall across sessions.
• Extensible Architecture: Supports Model Context Protocol (MCP) servers and a custom skill system with lifecycle hooks.
• Hardware Agnostic: Runs on everything from a Raspberry Pi to a Mac Mini or cloud servers.