Skip to content
kairo
Try Kairo

How Kairo uses AI

Last updated
Effective
Entity
Kairo Labs LLC · Illinois, USA

Last updated: May 2026

Kairo uses AI as connective tissue between your calendar, notes, tasks, and Spaces — never as the whole product. This page explains exactly where AI runs, what it sees, what we send to providers, and the limits we put in place to keep the system sustainable.

1. Where AI is used

  • Quick Capture — extracts events, tasks, and dates from text you type, paste, or drop in. Local heuristic by default; cloud parser on Plus/Pro for higher accuracy.
  • Note AI — summarizes, brainstorms, and answers questions about a specific note.
  • Document AI summaries — summarizes PDFs and long notes (Plus and Pro).
  • YouTube → notes (Pro) — generates a transcript, structured summary, and quiz from a pasted YouTube URL.
  • Smart Notes recording (Pro) — captures audio from your microphone for a lecture, meeting, or call; transcribes live and produces a rolling summary.
  • One-tap quizzes (Pro) — generates a personalized quiz from any note's content (multiple choice, fill-in-the-blank, short answer).
  • Academic Workspace tools (Pro) — syllabus extraction, flashcards, study guides scoped to a single course.
  • Kairo Copilot (Plus and Pro) — answers questions about your calendar, tasks, and notes in natural language.
  • Travel & commute intelligence (Plus and Pro) — leave-by alerts and route suggestions for events that need them.
  • Ambient hints — short, non-intrusive nudges on Home generated locally from your own data.

1a. What's gated by which plan

  • Free: Quick Capture (local heuristic), ambient hints, basic note summaries, limited daily AI calls.
  • Plus: + expanded AI capacity, document summaries, travel intelligence, conversational Copilot, larger context windows.
  • Pro: + YouTube → notes, Smart Notes recording, one-tap quizzes, syllabus extraction, study guides, highest AI throughput, the largest context window.

2. What we send to AI providers

Only the content needed for the specific feature you triggered. For Quick Capture, that's the text you entered. For Copilot, that's your question plus the relevant slice of events/tasks/notes — selected by a relevance and recency score, never your whole history. Long inputs are trimmed to a model-specific context budget before the call.

We don't send: passwords, OAuth tokens, billing info, the content of notes you haven't opted into, or content from other users in shared Spaces (unless those users are part of the same request, like a shared note's own content).

3. Model routing — cheap by default, premium when it matters

Kairo runs a tier-aware model router. Lightweight tasks (Quick Capture, short summaries, structured extraction) use a small fast model. Heavier tasks (long document analysis, study guides, advanced Copilot turns) route to a more capable model — but only for paid tiers and only when content size justifies it. The decision is deterministic and shown in the request log when you ask for it via a data export.

4. Which providers we use and what they can do with your content

Kairo Labs LLC routes AI requests through the following providers, all US-based and bound by enterprise data-processing terms:

  • Anthropic PBC (Claude family) — used for most reasoning, summarization, and Copilot turns.
  • OpenAI OpCo LLC (GPT family) — used for select tasks where it outperforms on capability or cost.
  • We may route some requests through Vercel AI Gateway for load balancing and observability; the underlying model providers remain the two listed above.

Zero data retention. We have configured zero-data-retention (ZDR) on our provider accounts where the provider offers it. Where ZDR is not available, we use the shortest retention window the provider offers (typically 30 days for abuse review) and the contractual prohibition on training applies.

No training on your data. Under our enterprise terms with Anthropic and OpenAI, providers are contractually prohibited from training their models on Kairo customer data, from selling it, and from sharing it with any third party for any purpose other than serving the request.

Smart Notes audio uses your browser's built-in speech recognition where supported; nothing leaves your device for that step unless you are explicitly using a cloud-only recorder. The complete list of all sub-processors is on the sub-processors page.

5. Rate limits, abuse protection, and account verification

Each plan has fair-use daily and monthly limits on AI generations to keep the platform sustainable. Limits are intentionally generous on the paid tiers and are designed to be invisible during normal use. We also run soft anomaly detection — short cooldowns after very rapid bursts or repeated identical inputs — to discourage automation farming. We do not ban accounts for normal human use. Heavy AI features (YouTube, Smart Notes, syllabus extraction, automations) require email verification.

6. Cache and reuse

For deterministic tasks (syllabus extraction from the same PDF, YouTube notes from the same URL, quizzes from an unchanged note), we cache the output for a bounded window so repeat requests don't pay the AI cost twice. Cached results never cross users unless the content itself is already shared inside a Space.

7. Your control

  • AI processing is off by default until you opt in.
  • Toggle it any time in Settings → Privacy & AI.
  • When off, Kairo falls back to local heuristic parsing — slower and less accurate, but no data leaves your device.
  • You can export your data and your AI request history any time in Settings → Data.

8. Limitations you should know

  • AI can be wrong. Confirm dates, times, and quiz answers before relying on them.
  • AI doesn't make autonomous decisions for you — every action requires your approval.
  • AI is not a substitute for professional advice. If a capture says "doctor told me to take X medication," Kairo parses the schedule, not the medical guidance.
  • Smart Notes transcription accuracy depends on your browser, microphone, and the room's acoustics.

8a. Not for high-risk uses

Kairo's AI features are for productivity and coordination. They are not designed, tested, or licensed for, and you must not rely on them for:

  • medical diagnosis, treatment decisions, or clinical workflows;
  • legal advice or substantive legal drafting;
  • financial, tax, investment, or insurance advice;
  • employment, housing, credit, education, or other consequential decisions about a person;
  • operation of vehicles, machinery, weapons, or any safety-critical system;
  • emergency or crisis response.

Output is informational only. You are responsible for independently reviewing and verifying any AI output before acting on it.

8a. Sensitive data — what NOT to send to the AI

Even with zero-retention contracts in place, treat any AI prompt as something that could leak. Never paste the following into Copilot, Capture, or any other AI surface:

  • Passwords, API keys, or session tokens.
  • Full credit card numbers, CVVs, or bank account / routing numbers.
  • Government IDs — Social Security numbers, passport numbers, driver's license numbers, national ID numbers.
  • Full dates of birth combined with any of the above.
  • Other people's medical, financial, or identity information unless you have explicit authorization to process it.
  • Material that you're bound by NDA, classification, or contract not to disclose.

Kairo does not scan prompts for these patterns and cannot retroactively unsend them. This rule protects you (from accidental leaks) and Kairo (from being a vector). If you do paste something sensitive, treat the conversation as exposed and rotate the credential.

8b. Opting out of AI features

AI processing is off by default. You can turn it on or off any time at Settings → Privacy & AI. Disabling AI will degrade or remove some features (Copilot, document summaries, YouTube notes, smart capture, study guides); core calendar, tasks, notes, and sharing continue to work without AI.

9. Automated decision-making (GDPR Art. 22)

Kairo does not make decisions that produce legal or similarly significant effects about you solely through automated means. AI output is advisory and acts on your own personal data only — your calendar, your tasks, your notes.

You control the autonomy level. By default Kairo is in review-required mode: AI-parsed items land on a review page and never enter your calendar or task list until you explicitly approve them. In Settings → Intelligence you can opt into auto-apply, where Kairo writes parsed items directly with a toast confirmation. Auto-applied items are flagged with a small “Auto” pip in the UI so you can always see what the AI decided versus what you decided. You can flip back to review-required at any time. Either mode, you own the data and can undo, edit, or delete anything.

10. Questions

Reach privacy@heykairo.io with anything AI-related — including data subject requests under GDPR Art. 13(2)(f) and CPRA.