XDA’s piece on giving Claude read access to an Obsidian vault surfaced something years of note-takers have suspected: the value of your notes compounds when a model can see across all of them at once. Your open tab in front of you can hold one insight; a model with vault-wide read access can trace a thought from a 2021 book review to a 2024 project retrospective in a single answer. It is the killer app for the plain-text-notes crowd.
Getting there safely is the tricky part. Most cloud AI tools want to upload your notes to their servers. Most local models are still too weak to spot the connections that make this worth doing. And the tooling around Model Context Protocol (MCP) and local RAG is moving fast enough that the “right” answer changes month to month.
We tested seven apps for AI-connected note vaults on desktop that keep the equation working: notes stay yours, the model gets the context it needs, and the setup is either free or reasonable. Every pick works on Windows, macOS, and (in most cases) Linux, and none of them force you to move your vault to a proprietary format.
What to look for in an AI-connected note vault app
Not every “AI note app” solves the same problem. The choices depend on where you want the AI to sit relative to your vault.
- Local vs cloud model access. Local models keep everything on your machine but limit answer quality. Cloud models are smarter but need a trust decision about what leaves your device.
- MCP support. Model Context Protocol has become the standard way for AI assistants (Claude Desktop, some others) to read local files and vaults. Apps that speak MCP integrate cleanly with modern assistants.
- RAG quality. Retrieval-augmented generation quality depends on chunking, embedding model, and re-ranking. Apps that get RAG right find better answers.
- Vault format ownership. Your notes should stay in an editable format you can open with any text editor. Markdown is the current safe answer.
- Read-only enforcement. For most workflows you want the AI to read your vault, not silently modify it. Apps that enforce read-only by default reduce surprise.
Quick comparison
| App | Best for | Free plan | Starting price/mo | Standout feature |
|---|---|---|---|---|
| Obsidian (Smart Connections, MCP) | Existing markdown vault owners | Free | $0 (Sync $4) | Plain-text vault plus growing MCP ecosystem |
| Logseq | Outliner users who want AI on their graph | Free | $0 | AI-friendly block-level structure |
| AnythingLLM | Anyone who wants a self-hosted AI vault UI | Free | $0 | Local RAG with any model, any file type |
| Cursor | Developers with a code-plus-docs vault | 2-week trial | $20 | Best-in-class AI for text and code combined |
| Continue | Free VS Code extension for AI over local files | Free | $0 | Bring-your-own-model AI over any workspace |
| Zed | Fast collaborative editor with AI | Free | $0 | Rust-native editor with built-in AI |
| Cline | AI coding-and-research agent inside VS Code | Free | $0 | Multi-step reasoning over your workspace |
The apps
1. Obsidian — Best for existing markdown vault owners
Obsidian is where most vault owners already keep their notes, and 2026 finally brought first-class AI. The Smart Connections plugin does local embedding and semantic search over your entire vault. The Obsidian MCP server (community-built) exposes your vault to Claude Desktop and other MCP-compatible assistants. The result: keep your existing vault, add an evening of setup, and you can ask Claude “what have I written about attention residuals” and get an answer with links to the actual notes.
Where it falls short: MCP setup is documentation-heavy for a beginner. Smart Connections indexing can take a while on large vaults.
Pricing:
- Free: The desktop app, Smart Connections, and MCP server are all free.
- Paid: Obsidian Sync at $4/mo if you want encrypted vault sync. Model API costs are separate.
Platforms: Windows, macOS, Linux.
Download: Obsidian official site
Bottom line: The default answer for anyone already keeping notes in Obsidian. If your vault is not already here, this is where to migrate.
2. Logseq — Best for block-level graph structure
Logseq stores notes as blocks in a graph, and blocks turn out to be a great unit of retrieval for AI. Community MCP servers expose Logseq to Claude Desktop, and the Logseq AI plugin does local-vector search inside the app. Because blocks have their own IDs, references and backlinks come along for the ride.
Where it falls short: Outliner-first workflow is not for everyone. AI plugin ecosystem is smaller than Obsidian’s.
Pricing:
- Free: Everything, including the desktop app.
- Paid: Optional sync tier.
Platforms: Windows, macOS, Linux.
Download: Logseq on GitHub
Bottom line: Best pick for outliner users who want AI on a block graph. Skip if the outliner format is a hard no.
3. AnythingLLM — Best self-hosted AI vault UI
AnythingLLM is a self-hostable desktop app that turns any folder into a searchable knowledge base for any LLM you plug in. It supports Ollama for local models, OpenAI/Anthropic/Gemini/Mistral for cloud, and does document ingestion for markdown, PDF, DOCX, and web pages. If your vault is not in Obsidian and you don’t want to migrate, point AnythingLLM at your notes folder and you have a chat UI over them within minutes.
Where it falls short: Chat UI is not a note-taking UI. This is a companion to your editor, not a replacement.
Pricing:
- Free: The desktop app.
- Paid: A Cloud tier exists but is optional.
Platforms: Windows, macOS, Linux.
Download: AnythingLLM official site
Bottom line: Best pick when you want a chat UI over any file-based knowledge base. Skip if a chat UI isn’t what you want.
4. Cursor — Best when notes and code live together
Cursor is a VS Code fork built around AI. If your “vault” is really a mixed repository of code, ADRs, design docs, and notes, Cursor gives you the best current AI reading experience over the whole thing. Composer and Chat both take vault-wide context, and the workspace-aware retrieval spots connections between a two-year-old design note and today’s failing test.
Where it falls short: Not really designed for pure prose. Subscription is not cheap.
Pricing:
- Free: 2-week Pro trial.
- Paid: Pro at $20/mo. Business tier higher.
Platforms: Windows, macOS, Linux.
Download: Cursor official site
Bottom line: Best pick when your vault mixes code and prose. Skip if you keep them separate.
5. Continue — Best free AI extension for any local workspace
Continue is a free open-source VS Code and JetBrains extension that lets you attach any AI model (local via Ollama, or cloud) to your workspace. Because it treats the workspace as a searchable context, you can point it at a folder of notes and ask questions, or edit note content, without leaving your editor.
Where it falls short: Requires an editor you already use. Not a note-taking UI by itself.
Pricing:
- Free: Everything. Open source.
- Paid: None from Continue itself. Model API costs still apply.
Platforms: Any OS that runs VS Code or JetBrains IDEs.
Download: Continue official site
Bottom line: Best pick for the “I already have a great editor, just give me AI over local files” case.
6. Zed — Best fast Rust-native editor with built-in AI
Zed is the fast, collaborative editor from the ex-Atom team, and it has grown a first-class AI assistant that operates over your open workspace. It handles multi-cursor prose editing and code with the same responsiveness. If Obsidian feels heavy, Zed with a plain markdown workspace is a solid alternative.
Where it falls short: Prose-writing features are thinner than Obsidian’s. No community plugin ecosystem yet.
Pricing:
- Free: The editor. AI assistant has generous free usage before requiring an API key.
- Paid: Optional Zed AI subscription.
Platforms: Windows, macOS, Linux.
Download: Zed official site
Bottom line: Best pick when performance matters and you want a modern editor over your markdown notes.
7. Cline — Best multi-step agent over your workspace
Cline (formerly Claude Dev) is a VS Code extension that runs a multi-step AI agent over your workspace. Unlike a chat that answers a single question, Cline can decide to read multiple notes, cross-reference them, and produce a synthesised answer with citations back to the source files. Because it runs inside VS Code, it works over any folder — Obsidian vault, plain markdown, or a repo.
Where it falls short: Multi-step agents cost more in tokens than single-shot chats. Read carefully what it plans before approving actions.
Pricing:
- Free: The extension. Open source.
- Paid: None from Cline. Model API bills go to your provider account.
Platforms: Any OS that runs VS Code.
Download: Cline on GitHub
Bottom line: Best pick when a single-shot question isn’t enough and you want the AI to do research over your vault. Skip if approvals-per-step slow you down.
How to pick the right one
If you already keep notes in Obsidian, install Smart Connections and the community MCP server and you’re done. No migration, best return on time invested.
If you keep notes in Logseq, do the same with its plugins. Same idea, different app.
If your notes live in loose markdown folders and you don’t want a note-taking UI change, run AnythingLLM or Continue.
If your “vault” is really a codebase with docs, use Cursor or Cline.
If you want a fast modern editor and are okay hand-crafting the workflow, use Zed.
Whatever you pick, treat “AI reads your vault” as a privacy decision, not just a productivity one. Keep the AI in read-only mode until you’re confident, and prefer local models for anything sensitive.
FAQ
Is it safe to give an AI access to my note vault?
Depends on the model and its transport. Local models (via Ollama) never leave your machine. Cloud models via API do, so review the provider’s data usage policy and keep sensitive notes out of the indexed folder. MCP connections to Claude Desktop route through Anthropic’s infrastructure but do not train models on the data by default.
What is MCP and why does it matter?
Model Context Protocol is an open protocol from Anthropic that lets AI assistants read local files, call local tools, and interact with local services in a structured way. It matters because it turns “attach a file” into “browse my whole vault,” which is where these workflows get interesting.
Which vault format works best for AI retrieval?
Plain markdown, one note per file, with reasonable heading structure. That format ingests cleanly into embedding pipelines and returns useful chunks. Rich-text formats (like docx) are harder to chunk well.
Can I keep an air-gapped setup?
Yes. Run Ollama locally, run AnythingLLM or Obsidian Smart Connections against it, and no data leaves your machine. Answer quality will be lower than cloud models, but the trade is a real one.
What is the cheapest way to try this?
Install Obsidian and Smart Connections (both free). Add Ollama and a 7B or 13B model for local embedding and retrieval. Total cost: your time. If you outgrow local models, wire in a cloud API for chat only.
