Lumilio Agent

Lumilio Agent is the conversational organizing assistant: describe what you want in natural language, and it works within your current account and authorized repositories. Every operation that changes organization requires precise confirmation, and the result is authoritative only when the server returns a receipt.

Typical use

Build an album from selected photos:

1. Select a batch of photos from your Repositories (the selection scope is shown next to your message);
2. Ask: “Put these photos into an album called Beach Holiday”;
3. Check the impact scope the Agent shows (target album, media count);
4. Click confirm — this only submits; the interface shows submitting → committed / rejected / failed / cancelled;
5. Only after the server returns the effect receipt is the operation applied.

Other real tasks: batch tagging (“add the tag Trip 2024 to these”), review results (“list duplicates among these”), organize (“group the selected media by theme”). Cancelling, stopping, or closing the conversation never commits interrupted changes.

What is shown next to your message

Each user message shows the mode, selection/viewing scope, and mentions. Dropped mentions are not silently ignored: mentions that cannot be resolved or exceed permissions are marked explicitly so you can correct them.

A result set is a working snapshot of the current query and permissions; new imports, deletions, or permission changes can make it differ from a later browse page. To keep a result, pin it to the Board — conversation memory is temporary; pins, the Board, media changes, and receipts are the persistent facts.

States: not configured, disabled, unreachable

• Not configured: no LLM provider or model chosen; the entry says so explicitly;
• Disabled: the administrator turned off “Enable Lumilio Agent”;
• Unreachable: configured but connection fails, the secret is invalid, or validation did not pass.

All three states are shown honestly. Core media management does not depend on the Agent — importing, browsing, and organizing keep working in every state.

Configuration and privacy

Administrators configure it under Settings → AI (see AI settings):

• The provider is chosen explicitly; no default fallback. DeepSeek/Ollama need a Base URL; remote providers need an API key;
• Verify connection validates the unsaved draft only; failure saves nothing. Switching providers requires providing the secret again;
• “Enable the Agent” and “Verify the provider” are two different actions.

The data boundary depends on the provider: with a local provider (e.g. Ollama) your questions and referenced metadata stay on your device; a remote provider receives your questions, referenced context, and tool results — whether they leave your device depends on its data-retention rules. Review privacy policy and fees before enabling.