moltis.org

What It Is

Moltis is a Rust-native, single-binary, self-hosted personal AI agent server. It is not just a chatbot wrapper, nor is it a SaaS product tied to a specific cloud provider. Instead, it runs as a persistent agent on the user’s own Mac, Linux, Windows, Raspberry Pi, or server. After configuring model provider keys locally, users can access the same assistant through channels such as the Web UI, Telegram, WhatsApp, Discord, Slack, Matrix, Nostr, and Teams.

Core Capabilities

On the AI side, Moltis supports both cloud and local models, including OpenAI, Anthropic, Gemini, DeepSeek, Mistral, Moonshot, OpenRouter, Ollama, LM Studio, local GGUF, MLX, and OpenAI-compatible endpoints. It also supports fallback, metrics, and local embeddings. Feature-wise, it includes long-term memory, vector plus full-text search, web browsing, code sandbox execution, natural-language cron, CalDAV calendars, voice input and output, Webhooks, MCP, Hooks, GraphQL, and JSON-RPC. Security is one of its main selling points: HTTPS by default, Passkeys, scoped API keys, an encrypted vault, SSRF protection, origin validation, human approval, and Docker/Podman/Apple Containers/WASM sandboxes.

Pricing and Trial

The main text does not disclose Moltis’ own pricing model. The page presents it as an MIT-licensed open-source project that can be deployed via an install script, Homebrew, Docker, Cargo, or GitHub Releases; it can also be quickly tried on DigitalOcean or Fly.io. Note that costs for third-party model APIs, cloud servers, voice services, and similar components are borne by the user.

Pros and Cons

Its strengths are lightweight deployment, relatively strong auditability, good privacy control, and broad coverage across multiple models, channels, and extension interfaces, making it suitable for building a truly “always-on” personal AI assistant. The drawbacks are that self-hosting, sandboxing, model keys, and messaging platform configuration are not especially friendly for ordinary users; output quality has no independent benchmark data and depends on the selected model; and information on commercial support, SLAs, and paid plans is also unclear.

Who It’s For and Access from China

Moltis is better suited to developers, security-sensitive teams, advanced users who want to run an AI assistant locally, and teams that need to integrate internal workflows through Webhooks, MCP, or GraphQL. The main text does not specify access conditions from China. Availability of the domain, GitHub, Docker images, various model APIs, and channels such as Telegram, WhatsApp, and Slack may vary in mainland China; payment information is also not disclosed. If access or model calls are restricted, users can consider using local Ollama/LM Studio, or compare alternatives such as OpenClaw, Hermes Agent, Open WebUI, Dify, and n8n + LLM.