What can the Help AI actually do?

Three things. (1) Answer questions about the app — how does voice matching work, why does a speaker show 'Unknown', what does Mark Internal Use Only mean, how do I send a story by email. (2) Answer questions about your own data — who do I meet most often, what did I promise Kyle last week, what's open with the Bambu Wallet project. (3) Propose structured actions you can confirm before they run — reprocess a story, open a feature, surface a specific quote, navigate to a person. Every action shows you exactly what will happen before it fires.

Is the assistant context-aware?

Yes. The ? button passes the current screen's context into the conversation — story id, speaker id, screen label — so 'Why does this speaker show Unknown?' resolves to THIS speaker, not a generic answer. The 'Asking about: X' chip at the top of the chat shows you exactly what the assistant knows. If you arrive from Settings → Help (no specific page), the context is 'general'.

Where does the assistant get its answers?

Two sources, in order. (1) A 54-entry knowledge base hand-written by the Bonfiyah team covering every major feature, gotcha, and FAQ. The assistant cites which KB entry it pulled from — tap the citation to open the full article. (2) Your own data, accessed through eight read-only tools (search transcripts, list speakers, get story metadata, who-did-I-meet, etc.). Your data is never shared across users; the assistant runs in YOUR cohort only.

What's the structured-action confirmation pattern?

When the assistant decides an action would help — say, you ask 'reprocess this story' — it proposes the action in a card with a Confirm / Cancel button. The card shows exactly what will happen: 'Reprocess story 2C371C2C with 7 chunks. This will re-run voice matching and may change speaker labels.' Nothing runs until you tap Confirm. Inspired by Claude's tool-use confirmation pattern; built so the AI can't take destructive action without your explicit OK.

How does this fit into Apple's 5.1.2(i) third-party AI rule?

Help AI requires AI consent before it can chat — same gate as every other Pro AI surface. Until you grant consent, tapping the ? icon shows the AI Consent screen explaining that questions get routed to Anthropic's Claude model. Once consented, every query and the relevant tool results travel to Claude through Bonfiyah's backend proxy (we never expose direct API keys to the device). Apple's November 2025 5.1.2(i) rule requires naming each AI provider before data leaves the device; the Bonfiyah consent screen does exactly that, with Anthropic named.

Is there a usage cap?

200 messages/day on Pro AI. A quota strip at the top of the chat shows how many you have left. The cap is generous enough that the assistant feels free to use — typical active users hit 5–15/day. The cap exists to defend against runaway loops and abuse, not to ration a paid feature.

Will Help AI work offline?

No. The backend proxy calls Anthropic's hosted Claude model; both legs require network. If you ask offline, the assistant tells you it's offline and asks you to try again — it does NOT serve a stale answer or pretend to know. Help AI is the one feature in Bonfiyah where the answer accuracy matters more than the convenience of always-on, so we hard-fail with a visible error rather than degrade silently.

v3.0.900 · Pro AI · NEW

The app you don't have to read the manual for.

Tap the ? icon on any screen. Ask anything. Get an answer grounded in your own conversations — not a generic LLM hallucination.

Other apps make you hunt through a help center. Bonfiyah's assistant sees the screen you're on, reads your transcripts, knows your speakers, and walks you through anything. When it wants to do something — reprocess a story, open a feature, surface a quote — it asks first.

Get the app See what it can answer

What you get at each tier · Updated 2026-05-30

Help that scales with your plan.

Free isn't a paywall here — it's a full static help library, on-device, no network needed. Pro adds the topics that only apply if you're using Pro features (iCloud sync, AI summaries, transcript email, ACP/Internal flags, speaker memory), plus one conversational AI advisory per day so you can see what the upgrade feels like. Pro AI keeps the conversational `?` icon open on every screen with 200 messages a day.

Help surface	Free	Pro	Pro AI
Static FAQ library 54-entry knowledge base · 5 categories · search + category browse	Baseline (~43 entries)	Baseline + Pro topics (~56)	Everything
Pro-feature topics iCloud sync, AI summaries, transcript email, ACP/Internal flags, speaker merge + retrain	—	✓	✓
Diagnostics One-tap recording / sync / summary checks	✓	✓	✓
Report a Problem Direct path to operator queue · auto-resolves on the next deploy	✓	✓	✓
On-page ? destination What opens when you tap the help icon on any screen	Static FAQ · ranked for this screen	Pro topics + AI advisory teaser	Full conversational AI · context pre-loaded
Conversational AI (daily cap) In-app Claude-powered assistant · cites the KB · proposes tools you confirm	—	1 advisory / day	200 messages / day
Data-grounded answers Assistant reads your transcripts via 8 read-only tools (search transcripts, list speakers, get story metadata, etc.)	—	Within the daily advisory	Every message
Structured tool proposals AI proposes actions — Reprocess this story, Open Settings, Surface a quote — you Confirm before they run	—	Within the daily advisory	Every message
Cross-screen context handoff Story id, speaker id, screen label auto-passed into the chat — assistant knows what you're looking at	—	Within the daily advisory	Every message

Free

Help that doesn't need a network.

The full static library ships in the app. Search by keyword, browse by category, run diagnostics, file a report. No daily caps, no AI, no surprises.

Pro

Pro topics + a daily AI taste.

Get the articles that only apply once you're using Pro features. One conversational AI advisory per day — screen-aware, your-data-grounded — to see what the upgrade actually feels like before you commit.

Pro AI

The app you don't read the manual for.

The `?` icon on every screen pushes a full conversational assistant. 200 messages per day. Reads your transcripts. Proposes structured actions you confirm. Cites the KB so you can verify every claim.

Three classes of questions it can answer

The assistant that actually has the context.

1. About the app itself

A 54-entry knowledge base hand-written by the Bonfiyah team. Citations link to the source article.

· "Why does a speaker show 'Unknown'?"
· "How does voice matching work across recordings?"
· "What does 'Mark Attorney-Client Privileged' do?"
· "How do I send a story by email with the right person redacted?"
· "What's the difference between Pro and Pro AI?"

2. About your own data

Eight read-only tools let the assistant search your transcripts, list your speakers, fetch story metadata, and run who-did-I-meet queries — all inside your cohort, never shared.

· "Who did I meet with most often this month?"
· "What did Kyle promise me about the Q3 timeline?"
· "Find the moment Dylan said the wallet pricing was wrong."
· "What's still open on the Bambu Wallet project?"
· "Summarize my last conversation with Sarah."

3. Structured actions it can run (with your OK)

When an action would help, the assistant proposes it as a confirmation card. You see what'll happen before it runs.

· "Re-process the Q3 Approach story" → confirmation card → tap Confirm
· "Open the Truth Layer view for Sarah" → deep-link card → tap Open
· "Surface the quote where Greg agreed to the term sheet" → result card with timestamp + tap-to-play
· "Navigate to Pre-Brief for the 3pm meeting" → deep-link card

Why it's not just "chat with your notes"

Context-aware. Cited. Confirmation-gated.

It knows the screen you're on. The ? button passes story id, speaker id, and screen label into the conversation. Ask "why is this happening" and "this" resolves to THIS recording — not a generic answer about recording in general. The "Asking about: <X>" chip at the top of the chat shows you what the assistant has the context for.

It cites its sources. Knowledge-base answers carry citation chips. Tap the citation and you land in the full article — not the chat's interpretation of it. Your-data answers carry direct quotes with timestamps that tap-play in the original recording.

It can't run anything without confirmation. When the assistant decides an action would help, it proposes a structured card with Confirm / Cancel buttons. The card shows EXACTLY what will happen. Nothing runs until you tap Confirm. Inspired by Claude's tool-use confirmation pattern; built so the AI can never take destructive action behind your back.

It tells you when it's offline. Bonfiyah's Help AI hard-fails with a visible "I'm offline — try again" rather than degrade silently. Most help assistants serve stale or invented answers when the network drops. We chose accuracy over always-on.

It respects your AI consent gate. Help AI is Pro AI–gated through the same consent screen as every other AI surface. Until you grant consent, tapping ? shows the consent explainer naming Anthropic as the AI provider — same Apple 5.1.2(i) compliance pattern Bonfiyah uses everywhere.

200 messages per day. Plenty.

A quota strip at the top of the chat shows how many messages you have left today. The 200-msg/day cap exists to defend against runaway loops and abuse — not to ration the feature. Typical active users hit 5–15/day; the cap rarely matters.

Quota resets at midnight in your local time. If you genuinely need more, email [email protected] — we'll talk.

Pro AI users get this on every screen.

New conversation memory app. New questions you didn't know you'd want to ask. The assistant that knows your data is one tap away.

See Pro AI pricing All seventeen other Pro AI features

Stay near the fire.

We send a short note every couple of weeks: a feature deep-dive, a real-world recipe, a peek at what's new. Unsubscribe anytime.

No spam. We use ConvertKit. See our privacy policy.