Pro AI feature
Action items, with everything attached.
Every task, pulled from every recording, with owner, deadline, source quote, and a tap-to-play link to the exact moment it was said.
Where Promise Tracker ends, action items begin.
Promise Tracker watches for commitments — the things people promise to do for each other in conversations. Action items is the broader category: anything that needs to happen as a result of the meeting, whether or not someone explicitly committed to it.
"We should look at the AWS bill before next quarter" is an action item. Nobody promised it. But it needs to happen, it needs an owner, and if you don't capture it in the moment it dissolves the second the meeting ends.
Pro AI's action-item extractor scans every recording for these — explicit assignments, implicit ones, and the "we should..." tasks that everyone in the room nodded at and then forgot.
What every extracted action item carries
The action itself
A short verb-led description: "Send the AWS cost breakdown to Maya," "Cancel the recurring marketing-tools subscription," "Email the legal team about the consent question." Specific enough to do, short enough to skim.
Owner
Inferred from speaker context. Self-assigned ("I'll send this") goes to you; assigned ("Mike, can you send this?") goes to Mike via your speaker library. Ambiguous ownership flagged for review.
Deadline
Normalized into actual dates. "By Friday" becomes the literal next Friday. "End of quarter" becomes the literal quarter-end. "Next week" becomes the next Monday-Friday range. Vague references flagged for confirmation.
Source quote
The exact 8-15 word quote that produced the action item, with timestamp link. Tap it to play the original audio. Useful when the action gets disputed three weeks later.
Confidence score
High-confidence items are added automatically. Lower-confidence ones (the "we should..." nodding-without-commitment kind) are flagged for your review rather than auto-tracked. Adjustable threshold in Settings.
Project / context tag
Inherits the Project tag from the recording. So action items from the "Acme Account" project are auto-grouped, separate from action items from the "Q4 Planning" project.
Where the items go
In the app, action items live in a dedicated tab — a single feed of everything that needs doing, sortable by owner, deadline, project, or source recording. Tap any item to see the full quote and audio context. Mark items done; closed items archive but don't delete (the source recording's history of "what got done" is preserved).
In v3.0, two new aggregation surfaces sit on top of the per-recording feed:
- Project Actions — every action item, decision, and "mine" filter aggregated across every story in a project, including its sub-projects. Hierarchical project trees supported. The full view of "what's open across this engagement," not "what's open in this one meeting."
- My Action Items — owner-scoped aggregator across the entire library. Pulls every action item assigned to you across every recording you've ever made. The "what should I be doing today" view that doesn't require you to open one project at a time.
Outside the app, items export as:
- Markdown / CSV — clean, portable, opens cleanly into Notion, Obsidian, Linear, a custom kanban. The default export.
- PDF — branded export with the Email Intelligence deep-link scheme baked in. Tap any action heading in the PDF to open Bonfiyah at the source moment.
- iOS share sheet — routes to Apple Reminders, AirDrop, Messages, and any iOS-installed third-party task app. The share sheet is what gives you Reminders today.
No direct integration with Things 3 or Todoist — the iOS share sheet handles those today. No web dashboard. No API access for third-party automation. The point is to capture, not to build a workflow product.
FAQ
What's the difference between Promise Tracker and action items?
Promise Tracker watches specifically for inter-personal commitments ("I'll send the proposal Friday"). Action items is broader: any task that needs doing as a result of the meeting, including unowned "we should…" intents. Most users use both — Promise Tracker for accountability across people, action items for personal task capture.
Where does the inference run?
Server-side, under a no-training commercial contract. Action-item extraction runs on Bonfiyah's backend via Anthropic's Claude — like the rest of the cross-recording Pro AI features. Your transcript text is processed for extraction, never used to train any model, and your raw audio is deleted from our servers within 7 days.
Can I edit action items after they're extracted?
Yes. Tap any item to edit the action text, change the owner, adjust the deadline, or delete entirely. Edits don't propagate back into the source recording — they just change how the item appears in your action-items feed and in your exported task tracker.
What about recurring action items?
If the same action item appears in multiple recordings (e.g., a weekly standup keeps generating "send the burn-down chart"), Bonfiyah groups them into a single recurring item rather than 12 duplicates. The recurrence pattern is editable.
See a real action-items feed
We'll send a screencap of a real (anonymized) action-items feed from a sales rep's week — the items, the owners, the deadlines, and the export-to-Things flow.
No spam. We use ConvertKit. See our privacy policy.