Vault Intelligence

Getting started with Vault Intelligence ​

Turn your Obsidian vault into an active research partner in five minutes: install the plugin, connect a model provider, and ask your first question.

1. Install Vault Intelligence ​

Vault Intelligence is published in the official Obsidian Community plugins directory, at https://community.obsidian.md/plugins/vault-intelligence:

  1. In Obsidian, go to Settings > Community plugins. If community plugins are currently restricted, select Turn on community plugins first.
  2. Select Browse, search for "Vault Intelligence", then select Install.
  3. Select Enable to finish.

The plugin starts building its search index as soon as it is enabled. The index runs entirely on your device and it is a derived cache that can be rebuilt from your notes at any time, so no configuration is needed before the next step.

2. Connect a model provider ​

Vault Intelligence splits its intelligence so you can put each part where you want it:

  • Search and connections always run locally inside Obsidian, on your device.
  • AI agents (the Researcher, Gardener, and Solver) and the semantic embeddings that power discovery use a model provider you choose: your own machine, a local server, or the cloud.

The quickest path to the full experience is a free Google Gemini key:

  1. Get a free API key from Google AI Studio.
  2. In Obsidian, go to Settings > Vault Intelligence > Connection.
  3. Paste the key into the Google API key field. Obsidian stores it in your device keychain and it is never synced to other devices.

Gemini is the recommended default: nothing else to configure, and it is the only provider with the Researcher's built-in web search and Python code execution. Not sure yet? The computing model explains why an API key is recommended and what actually leaves your device: only the short note excerpts the agent retrieves, never your whole vault.

Other providers ​

Vault Intelligence is provider-agnostic. The reasoning model (the agents' brain) and the embedding model (how notes become searchable vectors) are configured separately in settings:

  • Ollama: fully private reasoning on your own hardware, or hosted Ollama Cloud through the same connection. The guide also covers model choice and hardware tuning.
  • OpenAI-compatible endpoints: LM Studio, vLLM, OpenRouter, the OpenAI API, Anthropic's compatibility layer, and more.
  • Transformers.js for embeddings: small models that run entirely inside Obsidian, with no server and no API key — a zero-config path to private, offline semantic search.
  • Voyage AI for embeddings: high-quality cloud embeddings using a voyageai.com API key.

WARNING

Changing the embedding provider or model triggers a full vault re-embedding the next time Obsidian exits, so settle on your setup early to avoid repeating the initial index.

3. Your first chat ​

Let's see the magic immediately.

  1. Click the Vault Intelligence icon in the left ribbon and select Researcher: chat with vault from the menu that appears.

    The Obsidian sidebar highlighting the Vault Intelligence brain circuit icon

  2. Type this exact query (or pick a topic relevant to your notes):

    "What is the most interesting connection between my notes on [Topic A] and [Topic B]?"

    (Replace [Topic A]/[Topic B] with real topics from your vault, eg "Gardening" and "Recipes")

What just happened? ​

  • It read your notes: The agent found relevant files without you searching for keywords.

  • It reasoned: It didn't just copy text; it understood the relationship between the concepts.

  • It answered: You got a synthesised answer, cited with links to your source files.

    A chat response showing citations as clickable links

4. Where to go next ​

Now that you are connected, try a specific workflow:

To deepen your setup:

TIP

If you write in Chinese, Japanese, Korean, Arabic, or another logographic or right-to-left script, check the Agent language and Embedding chunk size settings (the latter lives in the Explorer settings). The plugin automatically optimises chunking for the higher information density of these scripts, which keeps semantic retrieval sharp.