Vault Operator - Obsidian Plugin

Vault Operator<br>Sebastian Hanke5k downloads

Real AI agent for your vault. Coworker, Copilot & thinking partner, that maintains your memory & knowledge, adapts to your workflows, uses plugins, skills & tools with full safety controls. BYOK & MCP

Add to Obsidian

Overview<br>Scorecard<br>Updates170

*]:md:h-full">Agentic AI operating layer for your vault.

You describe a task, it plans, searches, reads, writes, and reports back. Every action is visible. Every write needs your approval. Every change is undoable in one click.

Free. Open source. Local-first. Works with cloud models, with your existing ChatGPT or Copilot subscription, or fully offline with Ollama or LM Studio.

Documentation | Install from Obsidian | Community page

What people are saying

"Vault Operator might be the best Obsidian agentic AI plugin out there."<br>Nick, Buy Me a Coffee

"I've just discovered your wonderful plugin, which to me is way more than a simple plugin. It is a real harness inside Obsidian. That's awesome!"<br>arkham000, GitHub

"Vault Operator is one of the most interesting and powerful Obsidian plugins I've tried so far. The combination of agent functionality, vault access and document processing is particularly impressive."<br>Stapledon-de, GitHub

"Love your work with Vault Operator."<br>mikaljrue, Buy Me a Coffee

"Vault Operator plugin is exactly what I was looking for. The ability to plug in MCP, the support for various models and providers, the skills, and workflows. I am really looking forward to get my hands dirty. I am hoping I won't need to use VS Code + GitHub Copilot to help me manage my vault anymore."<br>Buy Me a Coffee supporter

"I have only just started, but this is real motivation to get back into Obsidian again."<br>hkocam, Buy Me a Coffee (translated from German)

What you get

A chatbot reads your prompt and answers. Vault Operator runs a loop: it picks an action, executes it against your vault, feeds the result back to the model, and continues until the task is done.

Capture sources with block-level provenance. Drop a PDF into the chat, get a source note where every key claim links back to the exact paragraph in the original.

Three-layer memory across sessions. Short-term session summaries, durable facts that survive resets, and a soul profile of how you write and how you want the agent to behave.

Find notes by meaning, not by filename. Local vector index, full-text keyword search, graph expansion through wikilinks, and a local cross-encoder reranker, combined with weighted RRF.

Build Word and Excel files, draft PowerPoint decks (PPTX in beta). Turn project notes into a DOCX, structured data into an XLSX, or meeting notes into a draft PPTX.

Run a vault health check. Surfaces orphans, broken links, missing backlinks, weak clusters, and over-connected hubs. Every fix creates a checkpoint you can undo.

Use the vault from ChatGPT, Claude Desktop, or Perplexity. Vault Operator runs as an MCP server, so your other AI clients can read the same memory and history as the in-Obsidian agent.

Hold the keys with auto-approve. Fail-closed by default. Per-category toggles for read, write, plugin-API, command, MCP, and web. Sensitive folders are gated by a .obsidian-agentignore file.

Reuse what Obsidian already exposes. Plugin-API discovery lets the agent invoke installed plugins (Excalidraw, Dataview, Tasks) instead of duplicating their work.

What it does for knowledge work

Capture sources with provenance

Drop a PDF or a Markdown source into the chat and ask for an ingest. The agent produces a clean source note with block IDs on every key claim, so each fact links back to the exact paragraph in the original.

Two paths:

/ingest for quick capture. Single-pass. One source, one note, about three minutes.

/ingest-deep for sense-making. A five-step guided dialog: triage and decision, output mode selection, deep ingest of the source, write the sense-making notes, set backlinks. Five to fifteen minutes for a real research paper.

Sense-making tutorial | Block-level provenance concept

Search by meaning, not by filename

A local vector index over your vault, plus full-text keyword search, graph expansion through wikilinks, and a local cross-encoder reranker. Ask "what do I know about X?" and the agent finds notes whose meaning is related, even when none of them contain the words you used.

The background analysis also surfaces note pairs that discuss similar topics without any wikilink between them, so you can spot connections you never wrote down.

Knowledge discovery guide

Build Word and Excel, draft PowerPoint (PPTX beta)

Turn project notes into a Word document, structured data into Excel, or meeting notes into a draft PowerPoint deck. DOCX and XLSX output is clean and reliable. PPTX is in beta: corporate template cloning is not supported in this version, so treat client-facing decks as a starting point and finish them by hand.

Office documents guide

Keep the vault navigable

The vault...