Skip to content

AI Insights

What AI insights are (and aren't)

AI insights are observations about your data, not financial advice. WDIAG's AI can notice patterns, flag anomalies, summarize your projection in plain English, and answer questions about your spending — but it doesn't give investment advice, can't predict the future, and should never be the basis for major financial decisions.

The AI layer is also entirely optional. Every feature in WDIAG — budgets, goals, debt plans, projections, calendar — works fully without any AI. If you don't configure a provider, insights simply don't appear. Nothing else breaks.

What AI can do

  • Anomaly detection — "Your dining spending in March was 3x your usual amount."
  • Pattern surfacing — "Your grocery spending increases by about 20% in November and December."
  • Forecast narratives — "Based on your current trajectory, you'll pay off the car loan 4 months early."
  • Q&A — ask questions in plain English: "How much did I spend on subscriptions last quarter?" or "Which category has grown the most in the last 6 months?"
  • Categorization assistance — suggest categories for transactions the rules engine couldn't match (routed through the HITL queue for your review).

Asking a question

In the Insights section, use the Q&A interface to type a natural language question. WDIAG passes your question along with relevant summarized data to the AI provider. The response is shown inline.

Rate limit: 10 questions per minute per household.

Generated insights

The system periodically generates insights automatically — anomalies, patterns, and forecast summaries. These appear in the Insights feed. You can:

  • Accept an insight (marks it as acknowledged, can be reviewed later)
  • Dismiss an insight (removes it from the feed)
  • Act on a suggestion via the HITL queue

How to interpret suggestions

All AI suggestions are routed through the Recommendation system and the HITL queue. The AI proposes; you decide. Nothing is applied automatically without your review.

Treat AI suggestions as a starting point for investigation, not a final answer.

Token budget

Every AI request costs tokens — units of text processed by the model. For cloud providers (Anthropic, OpenAI), tokens cost real money. WDIAG lets you set limits to prevent surprise costs.

See Token Budgets for configuration details.

Privacy and data sent to AI

What gets sent to an AI provider depends on your configured privacy level. The default for all cloud providers is generalizations_only — abstract patterns, no amounts, no merchant names.

See Privacy Levels for a full breakdown of what each level sends.

Setting up AI

See AI Providers for setup instructions.