AI assistant & actions

Ask “what did I decide about the launch date?” and get an answer that cites the Meeting notes it came from, or have AI fill in a Movie’s fields and suggest tags for a new Weblink. AI in Lar is optional, and nothing leaves your device until you approve it in a consent prompt. Read AI consent & privacy before you connect a provider.

Configure AI

A new workspace uses the AI service on the Lar server it is connected to, so you do not need a key of your own. On desktop, sign in to your account on that server first.

To use your own API key instead (often called BYOK, “bring your own key”), open Settings → AI provider on desktop:

Under Preferred provider, choose Anthropic, OpenAI, Gemini, or xAI. Requests are then billed to your account with that provider.
Paste the key into the API key field, choose Save, then choose Test.
Optional: under Activity, enter a daily token cap in the provider’s spending cap field and choose Save. The cap is an estimate, not a billing guarantee, and its count resets when the app restarts.

Lar seals each provider’s key in the operating system’s keychain. The app window never receives it.

The browser and mobile apps cannot hold your provider key, so they always send requests to the server’s AI service. If the server’s operator manages AI, Settings → AI provider shows Operator-managed AI; otherwise you can add a key that the server holds for you. The same consent prompt appears before anything is sent.

Use AI Chat

Open AI Chat with Ctrl+Shift+A, the AI chat button in the sidebar, or Open AI Chat in the command palette.
Ask a question or request a change. Type @ to mention a specific Object.
Choose a citation in the reply to open the Object it came from.

Lar assembles context on your device from the Objects you mention, from full-text search, from links between Objects, and, on desktop with semantic search turned on, from meaning-based matches.

With Anthropic as the provider, the assistant can also use workspace tools: search, open an Object, run a TQL query, and list backlinks. Tool-activity chips in the conversation show each lookup. Web search is separate and off by default; see Web search and tools.

Lar saves each conversation as an AI Chat Object when you send the first message. It then syncs, shares, and deletes like any other Object.

Run AI actions on an Object

Open an Object’s ⋯ menu and point to AI:

Generate title with AI: for an untitled Object that has body content.
Summarize with AI: for Types with a body.
Suggest tags with AI.
Suggest related notes with AI.
Fill fields with AI: for Types with text properties.

Other AI actions live where the content is:

In the body, open a block’s action menu and choose AI assistant… to rewrite, shorten, translate, explain, or summarize that block.
On an Image, choose Analyze to extract text, describe it, and detect its colors.
On an audio recording, choose Transcribe or Summarize. Transcription needs a provider that supports it, such as OpenAI.
On a Weblink, choose Analyze link for a summary, category, and tag suggestions.

AI output is a proposal. Review it before you apply it, especially property values and links. If no provider is ready, the AI submenu shows its actions dimmed with Set up AI… below them.

Turn on semantic search

Semantic search matches by meaning rather than exact words, so “car” can find a note about “vehicles”. On desktop:

Install and run Ollama on your computer.
Open Settings → AI provider → Semantic search and choose Enable.
Choose Reindex to build the index.

The embeddings stay on your device. Until the index is ready, search and AI context use full-text search and links between Objects. The browser and mobile apps do not have semantic search.

Add AI steps to Workflows

Workflows can use AI: Classify, AI: Extract fields, and AI: Summarize steps. An automatic Workflow needs standing consent for each kind of AI step it uses. Without it, a match waits as a suggestion for you to approve instead of sending content on its own.

Keep sensitive Objects out

Open an Object’s ⋯ menu, point to AI, and choose Exclude from AI. Lar then leaves the Object out of AI Chat context, semantic indexing, Workflow AI steps, and the MCP tools that AI agents use. A Type can also be excluded, and its Objects inherit the exclusion.

Keep out of automatic AI context is a softer option: automatic context skips the Object, but you can still @-mention it or pin it to a chat deliberately.

Exclusion does not retract content you already sent to a provider.

Next: AI consent & privacy explains the consent dialog, standing consent, and the activity log.