Question 1

Do I need a specific Obsidian plugin for OpenClaw?

Accepted Answer

No. Obsidian stores notes as plain markdown files in local folders, so OpenClaw can access them through the standard filesystem MCP server with no plugin required. A dedicated Obsidian MCP server adds structured access to tags, frontmatter, and internal links — useful but optional. The basic integration works with just a file path.

Question 2

Will OpenClaw modify or delete my notes?

Accepted Answer

Only if you grant write permissions through the MCP server configuration. The default setup is read-only — the agent can search and read your vault but can't change anything. If you want the agent to create new notes or append to existing ones, you enable write access explicitly. The agent template permission layer controls this precisely.

Question 3

How large can my vault be before performance degrades?

Accepted Answer

Without RAG, performance starts degrading around 100–200 notes because the agent tries to process too many files per request. With RAG, vaults of 5,000+ notes work smoothly because the retrieval layer only surfaces the 3–5 most relevant notes per query. The bottleneck shifts from vault size to retrieval quality — how well your notes are structured and tagged determines whether the right notes surface.

Question 4

Do my notes leave my machine?

Accepted Answer

Only the retrieved chunks are sent to the LLM as part of the prompt — the same way any conversation content is sent. Your full vault stays local. With a local LLM via Ollama, nothing leaves your machine at all. With a cloud LLM, only the specific note excerpts the RAG layer retrieves are included in the API request. The vault itself is never uploaded anywhere.

Question 5

What's the difference between Obsidian access and Supermemory?

Accepted Answer

Obsidian holds your explicit knowledge — the notes you've written. Supermemory holds the agent's operational memory — what it's learned from working with you across sessions. Your Obsidian vault tells the agent that your pricing is $X. Supermemory tells the agent that last time you discussed pricing, you preferred the quarterly billing framing. They're complementary layers, not alternatives.

Question 6

Can I use this with Apple Notes or Notion instead of Obsidian?

Accepted Answer

Obsidian is the simplest because it's plain markdown on your filesystem. Apple Notes uses a proprietary database and requires a Mac Mini to access programmatically. Notion works through its API but adds network latency and a cloud dependency. Both are possible with different MCP servers, but Obsidian's "just a folder of markdown files" architecture makes it the path of least resistance for local, private knowledge integration.

Question 7

Which Obsidian integration approach should I use — CLI Skill, MCP server, or plugin?

Accepted Answer

It depends on what you need. The filesystem MCP server is the simplest — raw read access, no extra setup. The OpenClaw Obsidian Skill (CLI) adds built-in vector store indexing, so your agent gets semantic search out of the box without configuring a separate RAG pipeline. The dedicated MCP server with RAG gives you the most control over embedding models, chunk sizes, and retrieval thresholds. A native Obsidian plugin bridges everything from inside Obsidian itself. For most users, starting with the Obsidian Skill or the dedicated MCP server is the right call — you get semantic search without building a custom pipeline.

Question 8

Does the MCP setup work with AI tools other than OpenClaw?

Accepted Answer

Yes. MCP is a universal standard, not an OpenClaw-specific protocol. Any MCP-compatible AI assistant can connect to the same server and access the same indexed vault. You configure the knowledge layer once — the RAG pipeline, the vector index, the vault mapping — and any compatible agent can query it. This means switching or adding agents doesn't require rebuilding your integration from scratch.

Question 9

How do I start if I don't have an Obsidian vault yet?

Accepted Answer

Start with 3 folders: Projects (active work), Resources (reference material), and Decisions (why you chose X over Y). Write atomic notes — 1 idea per note, 100–300 words, with clear titles. After 50 notes, connect the vault to OpenClaw via the filesystem MCP server and test with real queries. You'll see immediately which notes are useful to the agent and which need restructuring.

Tag: Obsidian

OpenClaw + Obsidian: Building a Knowledge-Connected AI Agent