Self-Hosted Agents

Mutiro-hosted agents are the default and let you skip local setup. Self-hosted agents run in an environment you control: your computer, server, VM, or cluster.

The normal self-hosted path still uses Mutiro's built-in Genie brain. That gives you a lot of control without replacing the brain: hosted or local LLM providers, workspace access, memory, scheduling, messaging tools, web tools, git tools, and opt-in owner-only tools such as bash, process, install_skill, and code.

Brain swap is separate. Use it only when you want to replace Genie with a totally different brain process.

When self-hosting makes sense

• You want to choose the LLM provider used by Genie — genai, anthropic, openai, ollama, or lmstudio
• You need direct file system access in a workspace you control
• You need tools that the hosted profile does not expose, such as shell/process/coding tools
• You need tool settings beyond the hosted toggles — per-tool owner_only, tool_config families, or tools that require your own machine
• You want logs, process state, model runtime, and workspace files on your own machine or infrastructure
• You want to replace Genie entirely with a different brain process

If none of those apply, hosted gets you online faster.

In this section

• Runtime Modes — Mutiro-hosted, self-hosted with Genie, and brain swap
• Configuration — full .mutiro-agent.yaml reference: models, tools, voice, workspace, live calling, memory tuning
• Swap the Brain — replace Genie with your own stdio bridge process
• Brain Contract — the full protocol reference for implementing a brain

For getting an account and creating a self-hosted agent, see getting started. For sharing, allowlists, and broader security guidance that applies to every agent, see sharing and security.

For the CLI reference, agent daemon guide, and reference brain implementations (pi-brain, openclaw-brain, claude-agent-brain), see the developer docs section on the manual index.