Vault Intelligence

Appearance

🗺️ Roadmap: Obsidian Vault Intelligence ​

Mission: To transform your Obsidian vault from passive storage into an active, intelligent partner that connects, verifies, and serves knowledge anywhere.

This document outlines the strategic direction for Vault Intelligence. It is a living document that evolves as we learn from our users and the rapidly changing AI landscape.

🟢 Phase 1: The Foundation (Completed) ​

Goal: Removing vendor lock-in and establishing a "Batteries Included" privacy-first experience.

  • [x] Universal "Batteries included" embeddings

    • The Vision: Install the plugin and it just works. No API keys required for basic search.
    • The Tech: We bundle a lightweight, high-performance model (like all-MiniLM-L6-v2) directly into the plugin. It runs 100% locally on your device, ensuring total privacy and zero cost.
    • Sovereign Intelligence: "Local-Only" switch that instantly cuts off all cloud API calls.

  • [x] "Sovereign intelligence" (user control & privacy)

    • The Vision: The "Right Model for the Right Task," transparently controlled by you.
    • The Tech: Smart routing between fast local models and powerful cloud reasoning.

🟢 Phase 2: The Agentic Revolution (Completed) ​

Goal: The AI takes initiative. It stops being a passive chatbot and becomes a proactive worker.

  • [x] "The Researcher" (deep reasoning agent)

    • The Vision: A dedicated agent that can read long documents (200k+ tokens) and perform multi-step reasoning.
    • Features: Auto-summarisation, cross-document comparison, and citation tracking using "Greedy Context Packing".

  • [x] "The Computational Solver" (code interpreting)

    • The Vision: The agent can write and execute Python code to analyse your data.
    • The Use Case: "Read my @Expenses note and forecast next month's spend."
    • Privacy: Code runs in a sandboxed WASM environment or via the Gemini Code Execution API.

  • [x] "The Gardener" (vault hygiene agent)

    • The Vision: An agent that proactively tidies your vault.
    • The Workflow: Scans recent notes, proposes an interactive plan, and applies changes safely after user review.
    • The Ontology: Introduces a formal structure (Concepts/, Entities/) so the AI knows where things belong.

  • [x] "The Explorer" (semantic navigation)

    • The Vision: A "See Also" sidebar that updates as you type.
    • The Tech: High-recall similarity search that finds related notes even if they don't share keywords.

🟡 Phase 3: Breaking Silos (Current Focus) ​

Goal: The Agent stops living in the sidebar. It works IN your editor and OUT with other apps.

  • [ ] "The Ghostwriter" (inline co-creation)

    • The Vision: Break the "Chat Sidebar" silo. The agent works directly in your editor, acting as a collaborative writer.
    • The Features: Inline Edit, Generative Insertion, and Smart File Creation.

  • [ ] Model Context Protocol (MCP) server

    • The Vision: Use your vault notes inside Claude Desktop, Microsoft Copilot, or other AI tools.
    • The Tech: Implement the MCP Standard to turn this plugin into a local server.

  • [ ] Multi-provider reasoning

    • The Vision: Freedom of choice. Use the best model for the job, regardless of who makes it.
    • The Tech: Abstraction layer allowing the Research Agent to run on OpenAI, Anthropic, or local open-weights models.

🟠 Phase 4: Visual Intelligence (The "Excalidraw" Stream) ​

Goal: Treating diagrams, sketches, and spatial layouts as first-class citizens.

  • [ ] "The Art Critic" (structure extraction)

    • The Insight: Standard search tools cannot see the relationships (arrows, groups, flow) encoded in drawing data.
    • The Tech: Parse compressed-json blocks to extract explicit connections.

  • [ ] ExcaliBrain graph reasoning

  • [ ] "Sketch-to-Structure" (de-rendering)

    • The Vision: Turn a messy whiteboard sketch into a clean note.

  • [ ] "Text-to-Diagram" (generative UI)

    • The Vision: Ask the agent to draw for you.

🔵 Phase 5: The Agentic Leap (Future Horizons) ​

Goal: Moving from "Questions" to "Tasks." The agent goes off, does work, and comes back.

  • [ ] Voice interface (desktop first)

    • The Vision: Talk to your vault while you work.

  • [ ] The "Analyst" (multimodal ingestion)

    • The Vision: Drag images, PDFs, and audio recordings into the chat.

  • [ ] Autonomous research reports

    • The Vision: Give the agent a job, not a prompt. "Research the current state of Solid State Batteries."

Phase 6: Blue sky (experimental) ​

Goal: Novel interaction paradigms that define the future of PKM.

  • [ ] "Graph Gardener" (maintenance agent)

    • A background agent that studies your vault's structure while you sleep.

  • [ ] Temporal intelligence ("vault evolution")

    • Analyse how your opinion on a topic has changed over time.

Technical architecture & challenges ​

1. The "Batteries included" embedding layer ​

  • Status: Delivered (Phase 1).
  • Next: WebGPU transition.

2. Editor integration ("Ghostwriter") ​

  • Constraint: Concurrency safety.
  • Strategy: Editor transaction API.

3. Model Context Protocol (MCP) implementation ​

  • Constraint: Local server security.

4. Handling Excalidraw hybrid files ​

  • Strategy: LZString decompression.

Contributing ​

This roadmap is not set in stone. We welcome community feedback!

  • Have an idea? Open a Feature Request.
  • Want to build it? Look for issues tagged help wanted or good first issue.

Research horizons (2026) ​

Experimental features targeting the new capabilities of Gemini 3, GPT-5, and Llama 4.

1. Visual vault indexing (multimodal RAG) ​

Index every chart, whiteboard photo, and PDF diagram.

2. Autonomous verification layers (corrective RAG) ​

The agent verifies its own retrieval quality before answering.

3. "Agent OS" orchestration (knowledge runtimes) ​

Treat the Vault as a "Knowledge Runtime" with specialized agents.

4. Federated RAG (privacy & silos) ​

Connect to data outside the Obsidian vault without importing it.