moltis-org/moltis

What it solves

Moltis provides a secure, self-hosted server for running persistent AI agents. It addresses the privacy and security concerns of using AI by ensuring that your data, keys, and agent execution remain entirely on your own hardware, while providing a unified interface to interact with multiple LLM providers and communication platforms.

How it works

Moltis operates as a single Rust binary that functions as a gateway server. It manages agent sessions, memory (via SQLite and vector storage), and tool execution. When an agent needs to perform a task, Moltis executes the command within a secure, isolated sandbox (using Docker, Podman, or Apple Container) to prevent host-level interference. It supports various communication channels like Telegram, Discord, and Slack, and integrates with external tools through the Model Context Protocol (MCP).

Who it’s for

This project is for power users, developers, and privacy-conscious individuals who want to run autonomous AI agents on their own infrastructure (such as a Mac Mini, Raspberry Pi, or private cloud server) without relying on third-party cloud-based agent platforms.

Highlights

• Secure by design: Features sandboxed execution, encrypted vault storage for secrets, and WebAuthn passkey support.
• Local-first architecture: A single, auditable Rust binary that eliminates the need for complex Node.js or Python runtime dependencies.
• Extensive integration: Built-in support for multiple messaging platforms, voice I/O, browser automation, and MCP servers.
• Memory and context: Includes hybrid vector and full-text search for long-term agent memory and project-aware context management.