Moltis – Self-Hosted AI Agent Server in Rust

TL;DR

TL;DR: Moltis is a self-hosted AI agent server written in Rust. One binary runs a persistent agent with sandboxed tool execution, multi-provider LLM support, and integrations for Telegram, WhatsApp, Discord, voice, and memory.

Source and Accuracy Notes

• Official site: https://moltis.org
• GitHub: moltis org on GitHub
• This post is based on the Show HN announcement and the official site.

What Is Moltis?

Moltis is a personal AI agent server that runs as a single Rust binary on your own infrastructure. Unlike cloud-based agent services, Moltis keeps everything in-house — your API keys, your conversation history, your integrations.

The core design is a persistent agent loop with sandboxed tool execution. Tools (functions the agent can call) run in isolated contexts, preventing a misbehaving tool from compromising the host system. The agent itself can connect to multiple LLM providers — OpenAI, Anthropic, local models via Ollama, and others.

Supported integrations include Telegram, WhatsApp, Discord, voice input/output, and a memory system for persistent context across sessions.

Setup Workflow

Step 1: Download the Binary

Moltis ships as a single self-contained binary. Download the latest release for your platform from the official site or GitHub releases page.

# Check the official site for the latest download link
# https://moltis.org

Step 2: Configure Providers

Create a moltis.yaml config file to specify your LLM provider and API keys:

llm:
  provider: openai  # or anthropic, ollama, etc.
  api_key: your-api-key-here
  model: gpt-4o

server:
  host: 0.0.0.0
  port: 8080

For local models:

llm:
  provider: ollama
  base_url: http://localhost:11434
  model: llama3

Step 3: Set Up Integrations

Enable integrations in the config:

integrations:
  telegram:
    enabled: true
    bot_token: your-telegram-bot-token
  discord:
    enabled: true
    bot_token: your-discord-bot-token
  voice:
    enabled: true

Step 4: Run the Server

./moltis serve

The agent starts listening on the configured port. Connect via Telegram, Discord, or the built-in web interface.

Deeper Analysis

Architecture: Moltis uses a persistent agent loop — the agent stays loaded in memory between requests, giving it memory of prior conversation context. Tools are invoked in sandboxed environments (separate processes or WASM contexts), providing a security boundary.

Multi-provider flexibility: The abstraction layer over LLM providers means you can swap GPT-4o for Claude or a local Ollama model without changing your tool definitions. This is useful for cost management or running entirely offline.

Memory system: Unlike stateless API calls, Moltis maintains a persistent memory store that the agent can query. This enables longer-running projects where context from weeks ago is still accessible.

Voice integration: Voice input/output is supported, making it viable as a personal voice assistant alongside the text-based interfaces.

Practical Evaluation Checklist

• One binary install — no Docker, no Node.js runtime required
• Sandboxed tool execution prevents tool-level privilege escalation
• Multi-provider LLM support with hot-swap between providers
• Persistent memory across sessions
• Telegram, Discord, WhatsApp integrations ready
• Voice I/O available
• Config file driven — no interactive setup wizard
• Open source (check GitHub for license details)

Security Notes

• Self-hosted means your conversation data never leaves your infrastructure
• Sandboxed tool execution limits blast radius of a compromised tool
• API keys stored in config file — apply standard file permissions (chmod 600 moltis.yaml)
• Review the GitHub security policy for vulnerability reporting

FAQ

Q: Can I run Moltis without an internet connection? A: Yes, if you use a local LLM provider like Ollama. Moltis works fully offline as long as your model server is local.

Q: What LLM providers are supported? A: OpenAI, Anthropic, Ollama (local models), and any provider compatible with an OpenAI-style API interface.

Q: How does the sandboxing work? A: Tools are executed in isolated contexts. The exact mechanism depends on the platform — check the GitHub repo for implementation details.

Q: Is there a cloud-hosted option? A: No — Moltis is designed to be self-hosted only. There is no managed cloud offering.

Conclusion

Moltis fills the gap between a stateless LLM API call and a full cloud agent platform. It runs on your own hardware, supports multiple LLM backends, and connects to the messaging platforms you already use. For developers who want a persistent personal agent without surrendering data to a third party, it is worth a look.

Visit https://moltis.org to get started.