Nexus gives AI agents like Claude full access to your Obsidian vault — so you can manage your entire knowledge base through natural conversation while keeping everything local.
Everything is natural language first. Notes, folders, workspaces, workflows, projects, tasks, search, image generation — the AI can create, read, update, and manage all of it just by you asking. There's also a settings UI for when you prefer clicking, but you never have to leave the conversation.
Nexus is the successor to Claudesidian. Legacy installs in
.obsidian/plugins/claudesidian-mcp/still work.
- Download the latest release:
manifest.json,styles.css,main.js,connector.js - Place them in
.obsidian/plugins/nexus/(or keep legacy.obsidian/plugins/claudesidian-mcp/) - Enable Nexus in Settings → Community Plugins
- Restart Obsidian
The built-in chat works right away with no extra software. Just add an API key in Settings → Nexus → Providers and start chatting inside Obsidian. Supports Anthropic, OpenAI, Google, Groq, Mistral, OpenRouter, Perplexity, Requesty, Ollama, LM Studio, Claude Code, Gemini CLI, Codex (ChatGPT), and GitHub Copilot.
Prerequisite: Claude Code must be installed on your computer and you must be signed in. Nexus uses your existing CLI installation — it won't work if Claude Code isn't installed and authenticated outside of Obsidian first.
- Install Claude Code and run
claudein your terminal to sign in - In Nexus, go to Settings → Providers → Anthropic
- Click the Connect button under "Claude Code"
- In chat settings, select a model labeled (Claude Code) — e.g. "Claude Sonnet 4.6 (Claude Code)"
Your messages route through the local Claude CLI using your existing subscription. Desktop only.
Prerequisite: Gemini CLI must be installed on your computer and you must be signed in. Nexus uses your existing CLI installation — it won't work if Gemini CLI isn't installed and authenticated outside of Obsidian first.
- Install Gemini CLI:
npm install -g @google/gemini-cli - Sign in: run
geminiin your terminal and complete the Google auth flow - In Nexus, go to Settings → Providers → Google AI
- The "Gemini CLI" section shows Connected once your credentials are detected
- In chat settings, select a Gemini CLI model
Your messages route through the local Gemini CLI using your existing Google account. Desktop only.
Connect your GitHub Copilot subscription via OAuth — no API key to manage.
- In Nexus, go to Settings → Providers → GitHub Copilot
- Click Connect
- A code appears in the modal — copy it, then enter it at the GitHub page that opens automatically
- Once authorized, Nexus fetches available Copilot models automatically
Requires an active GitHub Copilot subscription. Desktop only. Experimental — uses an undocumented Copilot proxy.
Connect your ChatGPT account via OAuth to access GPT-5 models through the Responses API.
- In Nexus, go to Settings → Providers → OpenAI
- Click Connect under "ChatGPT (Codex)"
- Sign in with your ChatGPT account in the browser window that opens
- In chat settings, select a model labeled (ChatGPT)
Requires an active ChatGPT Plus or Pro subscription. Desktop only. Experimental.
Nexus works as an MCP server with Claude Desktop, Claude Code, Codex CLI, Gemini CLI, GitHub Copilot, Cline, Roo Code, Cursor, Windsurf, and any other MCP-compatible tool. Requires Node.js (v18+) on your machine. The native chat inside Obsidian works fine without this.
Quick setup for Claude Desktop:
- Open Claude Desktop → Settings → Developer → Edit Config
- Add Nexus to the config (or use one-click: Settings → Nexus → Get Started → Add Nexus to Claude):
{
"mcpServers": {
"nexus": {
"command": "node",
"args": ["/path/to/Vault/.obsidian/plugins/nexus/connector.js"]
}
}
}- Fully quit and relaunch Claude Desktop
Setup guides for all supported agents →
For best results, give your AI agent some guidance on how to use your vault:
At the start of every conversation, list workspaces and load the appropriate one. If one doesn't exist, create one. If you lost context from compaction, reload the workspace. Always call getTools before useTools — never guess parameters.
Full recommended system prompt →
Just tell the AI what you want in plain language. Everything below works through Claude Desktop or the built-in native chat — no menus or special syntax required.
"Summarize my meeting notes from this week"
"Create a new note called 'Project Roadmap' with sections for Q1 and Q2"
"Move everything in my Inbox folder to the Archive"
Workspaces scope your AI sessions to a project. The AI creates them, loads them, saves states, and manages everything inside them — you just ask.
"Create a workspace called 'Thesis Research'"
"Save the current state as 'Literature Review Complete'"
Workflows are reusable routines that live inside a workspace. The AI can create them, bind prompts, set schedules, and trigger them — all through conversation.
"Add a daily workflow that summarizes notes I modified yesterday"
"Run the Morning Briefing workflow now"
Create projects, add tasks with dependencies, update status, assign work, set due dates — the AI handles the full lifecycle. There's also a settings UI if you prefer, but conversation is the primary interface.
"Create a project called 'Website Relaunch' and add tasks for design, implementation, and testing"
"What can I work on next?"
Find notes by what they're about, not just keywords. Search past conversations too. All local — no external API calls.
"Find everything I've written about behavioral economics"
"What did we discuss about the API design last week?"
Create reusable prompts with specific instructions, bind them to workspaces or workflows, and invoke them anytime. The AI can create, edit, list, and run prompts through conversation.
"Create a prompt called 'Code Review' that analyzes a note for bugs and suggests improvements"
"Run my 'Weekly Reflection' prompt"
In the native chat, type @ to quickly browse and invoke your saved prompts from the keyboard.
Select text in any note, right-click → Edit with AI, and transform it with a natural language instruction. Streaming preview, retry, and undo built in. Details →
Open the built-in chat (ribbon icon or command palette) to talk to any configured LLM with full vault access. Type / for tools, @ for custom prompts, [[ to link notes.
Details →
Install app plugins like ElevenLabs to add capabilities. The AI uses them just like any other tool.
"Convert my blog post note to speech and save it to Audio/"
See more examples of what you can do →
| Feature | Desktop | Mobile |
|---|---|---|
| Native Chat | Yes | Yes |
| Inline AI Editing | Yes | Yes (command palette) |
| MCP Bridge (Claude Desktop) | Yes | — |
| Local Providers (Ollama/LM Studio) | Yes | — |
| Semantic Embeddings | Yes | — |
| Cloud Providers | Yes (real streaming) | Yes (buffered) |
Nexus uses a two-tool architecture — instead of exposing 45+ tools upfront, it gives the AI just two: one to discover available tools, and one to execute them. This keeps things fast and works well even with smaller models.
All your data stays in a .nexus/ folder inside your vault as sync-friendly JSONL files with a local SQLite cache.
Architecture details → | Full tool reference →
- MCP server runs locally only — no remote connections
- All file operations stay inside your vault
- Network calls only for LLM providers you configure
- Embeddings download once, then run fully on-device
| Issue | Fix |
|---|---|
| Server not found in Claude | Settings → Nexus → Get Started → Add Nexus to Claude, restart Claude Desktop |
| Pipes not created | Make sure Obsidian is open (Windows uses named pipes) |
| Legacy install path | .obsidian/plugins/claudesidian-mcp/connector.js still works |
npm install # Install dependencies
npm run dev # Development build with watch
npm run build # Production build
npm run test # Run tests
npm run lint # Run ESLintMIT — see LICENSE for details.
Questions? Open an issue with your OS, Obsidian version, and any console logs.
