Moltis - Self-Hosted AI Agent Server in Rust

TL;DR

TL;DR: Moltis is an open-source Rust binary that turns any machine into a persistent AI agent server with sandboxed tool execution, multi-channel messaging (Telegram, Discord, Signal), voice I/O, and MCP integration — your keys never leave your hardware.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: moltis.org ← verified
• Source repository: github.com/moltis-org/moltis ← verified README
• License: MIT ← verified via GitHub API (tag shows MIT)
• HN launch thread: news.ycombinator.com/item?id=46993587 ← 131 points, front page
• Docs: docs.moltis.org ← verified
• Source last checked: 2026-07-20

What Is Moltis?

Moltis is a persistent personal agent server built in Rust. The README describes it as:

“One binary — sandboxed, secure, yours.”

Key claims from the official README:

• Single Rust binary — no Node.js, no npm, no runtime dependencies beyond the OS
• Sandboxed execution — Docker, Podman, or Apple Container; per-session isolation
• Multi-provider LLMs — OpenAI Codex, GitHub Copilot, and local model support
• Voice I/O — built-in STT and TTS providers
• Cross-channel — Telegram, Signal, Discord, Microsoft Teams, Slack, Matrix, Nostr
• Memory and recall — SQLite plus FTS and vector memory for hybrid search
• MCP support — stdio and HTTP/SSE MCP server integration
• Workspace size — approximately 270K lines of Rust across 59 modular crates

Setup Workflow

Prerequisites

• macOS, Linux, or a Raspberry Pi
• Docker, Podman, or Apple Container (for sandboxing)
• An LLM API key (OpenAI, GitHub Copilot, or a local model)

Step 1: Install the binary

macOS / Linux via install script:

curl -fsSL https://www.moltis.org/install.sh | sh

macOS / Linux via Homebrew:

brew install moltis-org/tap/moltis

Docker (multi-arch: amd64 and arm64):

docker pull ghcr.io/moltis-org/moltis:latest

Build from source:

cargo install moltis --git https://github.com/moltis-org/moltis

The Rust toolchain requires version 1.91 or newer.

Step 2: Start the server

After installation, open localhost:13131 in your browser. Paste your provider API key and select a model. OAuth providers like GitHub Copilot work with zero configuration.

Step 3: Connect a channel

Navigate to Settings → Channels in the web UI to connect Telegram, Discord, Signal, or other supported platforms.

Step 4: Connect an MCP server

Navigate to Settings → Tools → MCP to add stdio or HTTP/SSE-based MCP servers. The live tool inventory shows all available tools.

Architecture Overview

The Rust workspace is organized into 59 modular crates. Core crates from the official README:

| Crate | Role | |-------|------| | moltis-gateway | HTTP/WS server, RPC, auth, startup wiring | | moltis-tools | Tool execution and sandboxing | | moltis-providers | LLM provider implementations | | moltis-agents | Agent loop, streaming, prompt assembly | | moltis-chat | Chat engine, agent orchestration | | moltis-memory | SQLite + FTS + vector memory |

Security Model

The README calls out these security guarantees:

• Sandboxed execution — every tool call runs inside an isolated container
• Secret handling — API keys use secrecy::Secret, zeroed on drop, redacted from tool output
• Authentication — password, passkey (WebAuthn), and API keys with rate limiting
• SSRF protection — DNS-resolved, loopback/private/link-local addresses are blocked
• Artifact attestations — Sigstore keyless signing, SHA-256/SHA-512 checksums on releases

Verify a release with:

gh attestation verify <artifact> -R moltis-org/moltis

Comparison

Moltis positions itself against OpenClaw and Hermes Agent. Key differentiators from the README comparison table:

• Single binary vs. npm/TypeScript stack — no runtime dependencies beyond the OS
• Rust workspace — approximately 270K Rust LoC vs. approximately 152K Python LoC (Hermes Agent)
• Sandboxing — Docker/Podman plus Apple Container and WASM vs. local, Docker, SSH, Daytona, Singularity, Modal
• Vault — encryption-at-rest with XChaCha20-Poly1305 and Argon2id
• Skills system — bundled and workspace skills with autonomous improvement, plus OpenClaw import

FAQ

Q: Does Moltis require an internet connection? A: No. Once installed, Moltis runs entirely on your hardware. Your API keys never leave your machine.

Q: Can I use local models? A: Yes. The README lists local model support as a built-in feature alongside OpenAI Codex and GitHub Copilot.

Q: What platforms are supported? A: macOS, Linux, Raspberry Pi, and any server you own. The Docker image supports both amd64 and arm64.

Q: How does MCP integration work? A: Moltis supports both stdio and HTTP/SSE MCP servers. You configure them through Settings → Tools → MCP in the web UI.

Q: Is there a comparison with other agent frameworks? A: Yes. The README includes a full comparison table against OpenClaw and Hermes Agent.

Conclusion

Moltis fills the gap between single-user AI tools and cloud-hosted agent platforms. A single Rust binary handles the full stack — sandboxed tool execution, multi-channel messaging, voice I/O, memory with hybrid search, and MCP server connectivity. If you want an open-source, self-hosted alternative to cloud AI agents with verifiable security guarantees and no runtime dependency hell, Moltis is worth a look.

Start at moltis.org or read the full docs at docs.moltis.org.