Skip to content

pi-vault-mind

Passive Obsidian vault extension for the pi agent ecosystem.

Drop @agent-Miner, @agent-Broadcaster, or @agent-Manager markers in any Obsidian note. Save the file. The watcher detects the marker, groups by role, and dispatches isolated subagent forks. No chat pollution, no manual triggers.

Quickstart Install playbook Architecture

Why pi-vault-mind?

Traditional LLM chat sessions degrade as context grows. By task #12, the model is burdened by the irrelevant thought processes of tasks #1-#11. pi-vault-mind uses a multi-agent "Fork & Review" architecture to prevent that:

  1. You write in Obsidian. Drop @agent-Miner, @agent-Broadcaster, etc. in your notes.
  2. The Watcher dispatches. A background file-watcher detects markers, groups by role, and forks isolated subagents.
  3. Subagents execute silently. They write to LanceDB (vector + FTS + graph) and back to your vault, then terminate.
  4. You review the results. The dispatch IDs and synthesis notes show up in Vault/Agent/Inbox/.

No copy-pasting between tools. No terminal management. No context pollution.

Features

  • Passive File Watcher — drop @agent markers in any Obsidian note. The watcher handles the rest.
  • Multi-Agent Architecture — five specialist agents (Manager, Miner, Broadcaster, Heavy-Lifter, Watcher) with clean role separation.
  • LanceDB Vector + FTS + Graph — hybrid semantic + keyword search with automatic entity extraction and BFS graph traversal. All local, no external binaries.
  • Bidirectional Obsidian Sync — substantial entries auto-write to Vault/Agent/Inbox/. Graph entities render as Obsidian Canvas files.
  • Two-Layer Config — global (~/.pi/agent/pi-vault-mind.config.json) for shared settings, project (./pi-vault-mind.config.json) for project-specific knowledge.
  • Interactive Setup Wizard — /wiki setup walks through vault path, embedding provider, and model selection. CLI mode available for scripting.

Where to go next

  • Quickstart

    Install, configure, and run your first @agent marker in under 60 seconds.

    Quickstart

  • Install playbook

    The canonical 5-layer install (pi ext, skills, Obsidian, config, external CLIs).

    Install

  • Architecture

    Understand the "Fork & Review" model, the agent roster, and the dispatcher design.

    Architecture

  • Reference

    Tool, command, and configuration reference. Look up the exact behavior of any wiki_* tool or /wiki command.

    Reference

What people are saying

"The first Obsidian-LLM tool that doesn't try to be a SaaS. Local-first, agent-native, and the multi-agent dispatch is just plain faster than copy-pasting into a chat." — A user with 47 community plugins in their vault

License

MIT — see LICENSE.