Pro AI feature · v3.0
AI Project Context.
A 250–450 word briefing synthesised across every meeting in a project. Recurring participants. Open decisions. Pending action items. Themes. Watch items. Cached per project content signature, so it stays fresh on its own.
The "where were we" problem.
You haven't met with this client in three weeks. The project folder has six recordings in it. You have ten minutes before the call. Re-listening to six summaries is a non-starter, but you also don't want to walk in blind.
So you do what everyone does. You skim the most recent summary, hope it carries the load, and resign yourself to the moment in the meeting when someone references a decision you don't quite remember being on the call for.
Project Context is the feature that exists because that should be a 30-second read, not a 30-minute re-listen.
Real output
Six meetings. One brief. Read in 90 seconds.
Below is the actual Project Context output from a sample 6-meeting client engagement (names changed). The briefing is generated on demand and cached against a content hash of the underlying recordings.
Acme Co. — Q3 Migration
6 recordings · 4 participants · last updated 2 days ago
Acme is migrating their internal data warehouse from Redshift to Snowflake. The engagement began Feb 19; six recorded conversations span scoping, vendor evaluation, contract review, and a mid-project status check on Apr 22. Sarah Chen (VP Data) is the executive sponsor and decision-owner; Marcus Reyes runs day-to-day; Priya Nair is the technical lead and most consistent voice in the room.
The two open strategic decisions are whether to keep the existing dbt transformation layer (Sarah leans yes; Priya wants a clean rewrite) and whether the historical data retention window stays at 7 years (legal flagged a possible reduction to 5; unresolved as of Apr 22). Both have been raised twice without resolution.
Three commitments are pending. Sarah owes a Q3 budget revision — promised Apr 22, slipping since Mar 14. Marcus owes the updated DPA — promised week of Apr 8, no movement. Priya owes a final cutover plan, not yet due (May 15 target).
Two watch items: the team's anxiety around the cutover weekend has compounded across the last three meetings (specific quotes: "if Saturday goes sideways, Monday is unrecoverable", "we'd need to be ready to roll back by 9am"); and a recurring tension between Sarah's preference for a phased migration and Priya's preference for a single cutover that has not been resolved on the record.
Themes across recordings: rollback planning · vendor consolidation · post-cutover staffing · the dbt question
Generated in roughly 6 seconds the first time, served from cache after that. No manual refresh needed; the cache invalidates the moment the project's content signature changes.
How Project Context is built.
- 1
Hash the project's content.
A SHA-256 signature is computed across every recording in the project plus its sub-projects — transcript content, speaker assignments, edit history. That signature is the cache key. Two projects with identical content share a brief; a single edit to a single utterance invalidates the cache cleanly.
- 2
Resolve recurring participants.
Speakers are matched against your library across the project. The brief surfaces who shows up most — by airtime and by meeting count — and identifies the sponsor (longest tenure) and the tactical voice (most utterances per meeting). Persistent across the whole project, not just the latest recording.
- 3
Aggregate open decisions and pending action items.
Promise Tracker outputs and decision-shape utterances are pulled from every recording in the project and de-duplicated. A decision raised three times without resolution is flagged. A commitment that's been pending past its deadline is surfaced with its slip date — the brief reads "stalled since Mar 14", not "due Mar 14".
- 4
Surface themes and watch items.
A topic-clustering pass finds threads that recur across multiple recordings — the dbt question that's come up in four of six meetings, the staffing concern that's been raised quietly twice. Watch items are quotes from the actual transcripts, not paraphrases, so you can re-anchor on what was actually said.
- 5
Synthesise the brief.
A 250–450 word executive briefing — short enough to read in under 90 seconds, long enough to carry actual nuance. Written in the same dry, descriptive register as the AI Summary engine. No bullet-point soup; no filler. The brief is what a smart chief of staff would hand you in the elevator.
- 6
Cache, invalidate, never stale.
The brief is keyed to the content signature. Add a recording, edit a transcript, reassign a speaker — the cache invalidates the next time you open the project, and the brief is regenerated. You never see a stale brief. You also never wait when the brief is unchanged.
Who Project Context is for.
Anyone re-entering a project they haven't been in for two weeks.
Consultants + agencies
Walk into the client call already current.
You bill for the strategic conversation, not the catch-up. Project Context is the 90-second read that makes the next hour billable instead of remedial. One brief per client; refreshes itself when the engagement does.
Attorneys
Per-matter briefings, not per-meeting.
Treat the matter as the project. Every recorded call, intake, witness session, deposition — synthesised into a single defensible state-of-the-matter brief. Open decisions and pending commitments live at the matter layer, not buried in a single transcript.
Founders + execs
Re-onboard yourself into a workstream.
The fundraise, the hiring loop, the integration project — each one a project. The brief tells you what's open, who promised what, what's stalled, and what theme keeps surfacing. Read three of these and you've reconstructed your week.
Coaches + therapists
Per-client narrative across sessions.
Each client is a project. The brief surfaces what's recurring, what's resolved, what's been deferred — across months of recordings. Pre-session prep without re-listening to last week.
This is not a longer summary.
An AI Summary tells you what happened in one recording. Project Context tells you what's been happening across every recording you've made about a thing.
The shape is different. A summary is descriptive — it's the meeting you just had, summarised faithfully. Project Context is stateful. It carries forward unresolved threads. It tracks slips. It notices that a topic has surfaced four times without being decided. It surfaces the watch item that's been quietly escalating across three meetings without anyone naming it.
The other distinction: caching. A per-meeting summary is a one-shot. Project Context is keyed to a content signature, regenerates only when the underlying material changes, and is essentially free to re-read as many times as you want. The first read costs a few seconds of inference. Every read after that is instant.
Privacy
Where the inference runs.
The synthesis pass runs on a Bonfiyah-managed inference endpoint, not in a third-party LLM dashboard. Your transcripts are sent over a TLS tunnel, processed in-memory, and never written to log or training storage. The same posture as Promise Tracker and Story Mode.
The cache lives on your device. The content signature lives on your device. The brief itself is stored encrypted in your project folder and synced through your private iCloud — not through ours.
If you don't want any of this happening, the feature is opt-in per project. There's no default-on background synthesis.
FAQ
How long should a project be before Project Context becomes useful?
Three recordings is the practical floor. Below that, a per-meeting AI Summary is faster and roughly as informative. At three to six recordings, Project Context starts surfacing things no individual summary could — slips, recurring themes, unresolved decisions. Above six recordings, it's the only realistic way to stay current.
Does Project Context work across sub-projects?
Yes. A project can contain sub-projects, and the brief synthesises the entire tree by default. The hash key is computed across the full subtree, so adding a recording to a sub-project invalidates the parent's cache cleanly. You can also generate a brief for a single sub-project if you want a narrower scope.
Can I edit the brief?
No, by design. The brief is a function of the underlying content signature. If the brief is wrong, the place to fix it is in the underlying transcript or speaker assignment — and once you do, the cache invalidates and the brief regenerates. We resisted the impulse to make it editable because then it becomes another file you have to keep current by hand, which is the entire problem this feature exists to remove.
How do I share a Project Context brief with my team?
Export to PDF or Markdown from the project view. The PDF carries the brand chrome, the recurring participants list, the pending commitments with slip dates, and the watch items with quoted source text. The Markdown export is plain prose and pastes cleanly into Notion, Linear, or your CRM.
Does the brief include audio links?
Yes. Watch items, pending commitments, and decision references all carry deep links into the source recording at the source moment. In the in-app brief, those are tappable timestamps. In the emailed PDF, they're styled hyperlinks that open Bonfiyah and seek to the right second of the right recording — see the email-intelligence page for the format.
See Project Context on a real project
We'll send the actual brief above as a one-page PDF so you can see the formatting, the timestamps, and the watch-item quoting. No newsletter spam — just the artefact.
No spam. We use ConvertKit. See our privacy policy.