Mentat

Mentat - Obsidian Plugin

Your local agentic cognitive partner for Obsidian. Mentat is an AI agent that lives in your vault — it reads, writes, searches, researches, and summarizes your notes through conversation.

What you can do with it

• 💬 Chat with your vault: "Find my notes about RAG from last week and summarize them"
• 🔍 Deep research: Create structured research plans, search the web, fetch pages, and save findings
• ✍️ Write and edit notes: "Write a technical summary of this paper to Research/"
• 🛠️ Automate tasks: Bulk operations, frontmatter updates, file moves with link fixing
• 🎨 Two UI themes: Classic bubble chat or terminal-style timeline — switch anytime
• 🧩 Extensible: Extension system, skill framework, MCP support

AI Providers

Set up in Settings → Mentat → Add AI Provider.

Quick start

# Install
cd /path/to/vault/.obsidian/plugins
git clone https://github.com/Secbone/obsidian-mentat mentat
cd mentat && npm install && npm run build

# Enable in Obsidian: Settings → Community plugins → Mentat
# Configure: Settings → Mentat → Add AI Provider
# Open chat: Ctrl/Cmd+P → "Open AI Chat"

Development

npm run build        # Type check + bundle → dist/
npm run deploy:dev   # Copy to ~/Documents/obsidian/.obsidian/plugins/
npm run lint         # ESLint
npm test             # Vitest

Project structure

src/
├── agents/          # Agent runtime (BaseAgent, Compactor, events)
├── chat/            # Chat orchestration (session, query, subagents)
├── context/         # Message windowing, token estimation
├── extensions/      # Extension system (EventBus, ExtensionManager)
├── providers/       # AI providers (OpenAI, Anthropic, Ollama)
├── prompts/         # System prompts, skill prompts, vault context
├── settings/        # Plugin settings & settings UI
├── skills/          # Skill framework (registry, executor, MCP, strategies)
├── ui/              # Themed UI (ChatView, bubble theme, terminal theme)
└── types/           # Shared type definitions

In-depth docs

Roadmap

• Phase 1: Agent system — conversation, tool calling, multi-turn reasoning
• Phase 2: Skill framework — 12 built-in skills, progressive/native/auto modes, MCP
• Phase 3: Themed UI — BubbleTheme, TerminalTheme, smart scroll, terminal presets
• Phase 4: Extension system v1 + automatic context compaction
• Phase 4.5: EventBus architecture (AsyncGenerator → EventBus, AgentContext.confirmHandler removed)
• Custom skills: user-defined skills via .obsidian/plugins/mentat/skills/ with implementation.js
• Phase 5: Tree-structured sessions with branching and comparison
• Phase 6: Extension system v2 — dynamic loading, permission sandbox
• Phase 7: i18n, plugin marketplace, community extension registry

Built with

• Obsidian plugin API
• Anthropic Claude · OpenAI · Ollama
• PI Agent design influence (minimal core, extensibility)
• OpenCode design influence (event-driven agent, permissions)

License

MIT