gem-hunter
Language: TypeScript
Bring AI agents into Obsidian via Agent Client Protocol (ACP), such as Claude Code, Codex and Gemini CLI.
Chat with Claude Code, Codex, Gemini CLI, and any ACP agent right in Obsidian — mention your notes with @ and let the agent read and edit them in your vault. No copy-paste.
MCP servers, Agent Skills, slash commands, permission prompts — whatever your agent can do, it works here out of the box.
Built on the Agent Client Protocol (ACP) by Zed.
Side by side. Run several agents at once — in sidebar tabs, editor tabs, or floating windows — each with its own session and model. Broadcast a prompt to all of them, or cycle focus with a hotkey.
Any folder. Point a chat at any directory with New chat in directory — a code project, a writing project, wherever your agent normally works. The agent picks up that folder's context files and project configuration, and you can still hand it vault notes with @ mentions.
In your notes. Embed a live chat inside any note with a code block — pin an agent, and let the conversation survive restarts with persist. Or add one-click agent buttons that fire a prepared prompt.
Mission control. The Session Manager lists every open conversation across sidebar, tabs, floating windows, and notes, with live status icons — including "waiting for permission". Click any entry to jump straight there.
Any ACP agent. Nine presets — Claude Code, Codex, Gemini CLI, Mistral Vibe, OpenCode, Kiro, Hermes Agent, Pi, Grok Build — plus any ACP-compatible agent as a custom entry. New agent ships ACP support tomorrow? Add it as a custom entry — no plugin update needed.
@ — fuzzy search over names, paths, and aliases. The active note is auto-mentioned, down to the exact lines you selected[[wikilinks]], $LaTeX$, and clean Markdown tables (configurable)[[links]] inside mentioned notes are surfaced as paths so the agent can decide which ones to read/ commands, with inline argument hintsInstall and authenticate an agent, following its setup guide:
Claude Code · Codex · Gemini CLI · Mistral Vibe · OpenCode · Kiro · Hermes Agent · Pi · Grok Build · Custom Agents
Open Settings → Agent Client and check the agent's path — Auto-detect usually finds it
Click the robot icon in the ribbon and start chatting
Agent Client is a desktop-only plugin. It launches your locally installed agents as child processes and lets them run terminal commands — that is the core of what it does. The filesystem is only probed read-only for the Auto-detect button; all note reading and writing goes through Obsidian's vault API.
Keep in mind that the agents themselves have the same full system access they have in a terminal. The plugin surfaces every permission request so you can approve or reject each action; Auto-allow permissions (off by default) skips those prompts entirely — enable it only if you understand the implications.
API keys are stored in Obsidian's Keychain, never as plain text. What leaves your machine is what you send to your agent's provider: your messages, mentioned notes, and attachments.
To try pre-release versions before they are published to Community Plugins:
https://github.com/RAIT-09/obsidian-agent-clientmain.js, manifest.json, styles.css from ReleasesVaultFolder/.obsidian/plugins/agent-client/npm install npm run dev
For production builds:
npm run build
Apache License 2.0 — see LICENSE for details.
Thanks to everyone who has contributed!