Munin docs

Technical documentation for the local AI assistant.

Read how Munin boots, installs models, stores data, runs the chat pipeline, recalls memory, searches the web, executes tools, and manages workspaces.

Overview

The high level architecture of Munin and the boundaries between UI, services, storage, and native integrations.

How Munin starts, resolves the first destination, initializes services, and exposes navigation.

Model bundle installation, local inference, utility generation, performance profiles, and download recovery.

How a user message becomes a local response, including routing, memory, search, streaming, tools, and persistence.

Local vector recall, durable profile facts, memory management, and how memory enters the prompt.

Search routing, DuckDuckGo query execution, result parsing, page text extraction, grounding, and source handling.

Skills, tool fences, native intents, MCP server configuration, permission gates, and tool execution.

Workspace planning, review, research, execution, storage, scheduling, logs, and generated artifacts.

Voice mode recording, local transcription, chat handoff, streamed speech output, wake word handling, and cleanup.

What Munin stores, where it stores it, what reaches the network, and how destructive actions behave.

User-facing controls, permission surfaces, defaults, and how settings map to app behavior.