Skip to main content

Configure an AI provider

Enable the optional AI assistant and connect a language-model provider. For common cloud providers you pick from a suggested model list and paste an API key — you do not need to type model names, endpoints, or context sizes unless you want to.

Prerequisites​

  • Big Picture Money is unlocked (see Unlock the app)
  • An internet connection the first time you click Refresh model list (and whenever you refresh again)
  • For cloud providers: an API key from OpenAI, Anthropic, or Google Gemini (or another service if you use OpenAI-compatible)
  • For Ollama: Ollama installed locally with at least one model pulled (for example llama3.2)

Steps​

Load suggested models​

  1. In the left navigation, click Settings.
  2. Click AI assistant in the left list.
  3. Turn on Enable AI.
  4. Click Refresh model list. The app downloads the current provider and model suggestions from Big Picture Money’s servers and keeps them on your computer for later visits.
  5. After a refresh, if your selected model changes to a newer recommended model, a short notice appears under Model (for example Updated to the recommended model.).

Cloud providers (OpenAI, Anthropic, Google Gemini)​

  1. Choose a Provider (OpenAI, Anthropic, or Google Gemini).
  2. Choose a Model from the suggested list. The recommended (preferred) model is selected for you.
  3. Paste your API key. Next to the label, click create API key to open that provider’s key page in your default browser.
  4. Click Save if needed (some fields save when you leave them).
  5. The first time you send a message with transaction or screen context, read the data disclosure and click I understand to continue.

Your API key is stored in your operating system’s secure credential store, not in the database.

Custom model or advanced settings​

  1. In the Model list, choose Custom…, or expand Advanced.
  2. Under Advanced, you can type a Model name, Endpoint URL, and Context window (tokens) — how much transaction and screen data is included in chat (older context is summarized when over the limit).
  3. Click Save.

Ollama or OpenAI-compatible​

Ollama and OpenAI-compatible endpoint appear at the end of the provider list. Advanced opens automatically for these options.

  1. Choose Ollama or OpenAI-compatible endpoint.
  2. Under Advanced, set Model name and Endpoint URL (Ollama defaults to http://localhost:11434 when blank or as suggested).
  3. For OpenAI-compatible, paste an API key if the endpoint requires one.
  4. Optionally adjust Context window under Advanced.
  5. Click Save.

Azure OpenAI and similar hosted endpoints that need your own URL work best as OpenAI-compatible.

Accepting or dismissing category suggestions​

When AI suggests a category for an unknown transaction:

  1. Open Review queue or select the transaction in the register.
  2. Review the suggested category and explanation.
  3. Click Accept to apply the category, or Dismiss to leave the transaction unchanged.

Expected result​

  • Before the first Refresh model list, suggested models are not available; after refresh, the Model list shows suggestions for the chosen cloud provider.
  • Settings shows your chosen provider and model after save without requiring typed endpoint or context values for OpenAI, Anthropic, or Google Gemini.
  • create API key opens the provider’s key page in the default browser for those cloud providers; it is not shown for Ollama.
  • Choosing Custom… or Ollama / OpenAI-compatible shows Advanced with model, endpoint, and context window fields.
  • After Refresh model list, the model may update to the recommended model and a notice appears under Model when it changed.
  • With AI enabled, the right panel shows the AI assistant below transaction details (when the panel is open), and Chat in the left navigation opens the full-screen conversation page.
  • A local Ollama endpoint returns a chat reply within a few seconds.
  • A valid cloud API key allows chat without errors.
  • The panel header shows the active provider and model name.
  • Quitting and reopening the app preserves your AI settings and the last downloaded model list; API keys remain in secure storage.
  • Turning Enable AI off disables the assistant; you do not choose a “None” provider.
  • Sending the same unknown merchant description twice does not trigger a second LLM call for categorization (cached locally).
  • With AI disabled, the right panel shows only transaction details — no chat block.