Ratel
approvedby golddream-y
AI assistant for your notes—chat with your vault, semantic search with citations, multi-step research and safe edits. Works with DeepSeek, Claude, and Ollama. Local embeddings, no telemetry. - This plugin has not been manually reviewed by Obsidian staff.
Ratel
The graph-native AI agent for your vault.
Chat, research, and act on your linked notes. Ask what you wrote about a topic; have Ratel pull sources into a draft. Answers cite notes you can open in one click.
Why Ratel
- Graph-native — built for linked notes, not generic chat over files
- Fusion retrieval — semantic + keyword multi-way recall, automatically pulling in linked related notes; numbered citations open in one click
- Highly extensible — built-in skills & subagents, plus MCP servers (HTTP or stdio) to plug in web search and other external tools
- Deeply customizable — swap chat / embedding / rerank models, override any prompt section, choose your MCP ecosystem
- Private & safe by default — local embeddings; network only to your configured model API and the MCP servers you add; every vault edit asks your approval first
Install
Obsidian → Settings → Community plugins → Browse → search Ratel → Install → Enable.
Requires Obsidian 1.13.0+, desktop only.
Then: open Settings → Ratel → Chat model, pick a scene preset (DeepSeek / Ollama) or custom Base → wait for the first index → click the 🦡 ribbon (or run Ratel: Ask vault).
Full walkthrough: User Guide.
What you can do
Ask with citations
“What did I write about performance tuning?” — numbered [1][2] sources, click to jump.
Multi-step work
“Pull product-planning notes into a background doc” — search, read, synthesize, write (with confirmation before edits).
Know the room
Every turn injects local time. “Summarize this note” uses the active file. Daily note path, recent edits, and outlines are first-class tools.
Remember & extend
Say “remember I prefer Tailwind…” — stored as Markdown under .ratel/memory/. Drop a SKILL.md into .ratel/skills/ to teach new workflows, or attach an MCP server (Tavily, Brave…) to give the agent web search and other external tools.
Privacy
- Index and default embeddings stay on your machine
- Network access: by default only the model API you set (DeepSeek / Claude / Ollama)
- MCP servers: only the MCP endpoints you explicitly add in settings receive requests; each tool call asks your approval by default
- No analytics, no phone-home
Docs
| Doc | Contents |
|---|---|
| User Guide | Setup, scenarios, slash commands, FAQ |
| Changelog | Full release history (ranges OK for patch bursts) |
| Architecture | Ports, agent loop, tools, workers |
Issues & ideas: GitHub Issues.
License
For plugin developers
Search results and similarity scores are powered by semantic analysis of your plugin's README. If your plugin isn't appearing for searches you'd expect, try updating your README to clearly describe your plugin's purpose, features, and use cases.