Sage Chatbot

Overview

Sage Chatbot lets you interact with your finance data using natural language.

Use Sage for:

• Quick summaries.
• Directed questions about spending behavior.
• Guided discovery of related app workflows.
• Faster navigation from question to action.
• Voice-first queries when you prefer speaking.

How Sage routes requests:

• On-device AI is used first when available.
• Optional cloud AI (your own Gemini/OpenAI key) is used only when local AI is unavailable.
• If neither is available, Sage uses Basic mode.

Voice experience:

• Tap mic to start voice capture.
• A permission explainer is shown before microphone access is requested.
• On-device voice mode is currently active; Whisper cloud mode is shown as preview and is not enabled yet.

Best practice:

• Ask focused questions with clear time scope.
• Validate important outputs using linked screens.

Prompts & Suggestions

Prompt suggestions help you discover what Sage can do without guessing commands.

How to get better responses:

1. Mention period/time frame.
2. Mention account/category if relevant.
3. Ask one clear question at a time.
4. Use follow-up prompts to refine.
5. For voice, speak short single-intent commands first.

Examples of useful prompt style:

• “Show my top spending categories this month.”
• “What changed vs last month?”
• “Where can I reduce recurring costs?”
• “Add 500 for lunch from my last used account.”
• “Open upcoming dues.”

Avoid:

• Very broad multi-part prompts in one message.
• Acting on high-impact conclusions without validating source data.

Privacy & Limits

Sage behavior depends on enabled settings and available on-device/app data context.

Cloud AI privacy model:

• MoneySplit does not run its own AI servers for Sage cloud requests.
• Requests are sent directly from your device to your selected provider.
• Only a minimal structured context is shared for fallback routing:
  • locale and currency
  • sanitized account labels
  • upcoming pending count
  • active subscription count

Data not shared with cloud AI:

• Transaction rows
• Merchant names
• Category spend breakdown
• Raw SMS bodies
• Statement PDF text
• Transaction notes
• Full account numbers
• OTP or personal messages

Important limits:

• Response quality depends on data quality in transactions/categories/tags.
• Missing or poorly categorized records reduce answer reliability.
• Some actions may require manual confirmation in source screens.
• Cloud provider errors can occur due to invalid keys, unavailable model access, or provider quota/billing limits.

Responsible usage:

• Use Sage for acceleration, not blind automation.
• Validate critical conclusions in Analytics/Transactions before major financial decisions.