Everything Brieflee does, and how to drive it.

Brieflee listens to a meeting and works on it in real time. A notetaker joins the call as a visible participant and the transcript streams back, so the report is written from what was actually said rather than from memory.

During the call: a side panel, one only you can see, fills with a suggested reply, the next question worth asking, risks it heard, and what you still have not covered. Switch coaching on from the panel at any minute; the recording and the report happen either way.

It never speaks, never posts to the meeting chat beyond its own arrival notice, and never renders on the shared screen or in the recording. You stay the only voice in the room.

There are three ways in. The meeting app puts the panel inside the call. The web app lets you send the notetaker into any call from a link: no install on the meeting platform at all. The desktop app, for Windows and Mac, records calls on your own computer, with nothing joining them.

Microsoft Teams

Google Meet, Zoom, or anything with a link

Open the web app, press Send Brieflee to a meeting in the sidebar, and paste the join link. Brieflee validates it and sends the notetaker in. Recognised links:

In Zoom specifically, opening the Brieflee panel dispatches the notetaker on its own using the join URL the Zoom SDK provides.

Desktop app for Windows and Mac

On Windows 10 and 11, it notices when Teams, Zoom, Google Meet or Webex takes your microphone and records the call from your computer: your microphone and what the computer plays, streamed to Brieflee as you talk. When a Brieflee notetaker is going to the same booked meeting, the app leaves the recording to it, and records instead if the notetaker is turned away or still waiting to be let in after three minutes. Google Meet is recognised from its browser tab while that tab is in front; after that you can switch tabs and it stays one recording.

On a Mac with Apple silicon (M1 or later) and macOS 13 or later, it records the same two sides, with some differences. It starts by itself for a call in the Zoom, Teams or Webex app; for Google Meet or any other call in a browser, press Record. A Zoom recording ends with the meeting, but press Stop when a Teams or Webex call ends: a Mac does not tell Brieflee that the call is over, and left alone the recording carries on until nothing has been heard for five minutes. Installing on a Mac has the steps and the rest of what differs.

On either, nobody sees a notetaker join when the computer records, so tell everyone on the call that an AI notetaker is running. Download it, check its checksum, and find the MSI for IT deployment on Windows at brieflee.ai/download.

Seven tabs. Four sit in the bar: Reply, Feed, Probe, AI, and Risk, Gaps and Decided open from the “More ▾” dropdown at the end. Each carries a live count badge; Risk adds a red ! when a flag is high or critical.

The header shows the meeting title, an elapsed clock, a pulsing Rec light once the first line lands, and a running count of captured transcript lines. The clock is anchored to when the meeting actually started, server-side, so switching Teams tabs or reloading the panel does not reset it to zero.

A signal is something the AI produced: one of the four below. It is not the line count in the panel header, which counts lines of speech the notetaker finalised. The two used to share the word and a customer, quite reasonably, asked which was which.

If the client speaks again while a reply is still writing, the stream is cancelled and the half-written text is relabelled superseded rather than left looking finished. Below the live reply sit refined answers from a slower pass: the detected question as a pull-quote, a suggested reply, bullet talking points, a confidence label and a copy button.

Every meeting is tracked against a fixed ten-point technical checklist, shown as a meter with an NN / 10 count:

An item is never marked covered or partial without an actual quote from the call; if the AI claims coverage with no evidence, the server forces it back to empty. Expand a row to read the quote it was based on. You can override any item by clicking its status dot to cycle empty → partial → covered, and the change syncs to every device watching that meeting.

Set the expected meeting length (default 60 minutes) and the panel runs a T-minus countdown. With ten minutes left and items still open it writes a single bundled wrap-up question covering the remainder; at two minutes it switches to a red “These will need a follow-up” banner.

The Feed tab is the live transcript. Each finalised line carries a timestamp, the speaker's name as the notetaker reported it, and the text; every distinct speaker keeps a consistent colour and your own lines take the accent. A greyed “live” line can show a sentence forming before it settles.

Right-to-left scripts (Urdu, Arabic, Hebrew) are laid out correctly beside a Latin speaker name, both in the panel and in the finished report.

The AI tab answers questions about the meeting in progress from the transcript so far, and is instructed to say plainly when the transcript does not cover something rather than guess. The conversation survives a panel reload. In-meeting questions are free.

The notetaker is a browser bot running on Brieflee's own infrastructure. It joins as a guest participant named Brieflee Notetaker, visible to everyone in the participant list. Nothing is captured covertly.

On joining, it posts this to the meeting chat:

To help take notes, this meeting is being recorded and transcribed by Brieflee AI. By staying in the meeting, you're giving consent. A summary may be shared with the meeting organiser. Privacy policy: https://brieflee.ai/privacy

By default no meeting audio is sent to a third-party speech engine at all: the bot turns on the meeting platform's own closed captions and Brieflee ingests the caption text, with speaker attribution coming from the platform itself.

You pick the spoken language of the meeting per account, and it is read at the moment the bot is created, so it applies identically to a notetaker you send in by hand and one dispatched from your calendar.

Connect Outlook or Google Calendar and the notetaker turns up on its own, dispatched1, 3 or 5 minutes before each meeting starts (3 unless you change it under Calendar auto-join in Settings, Integrations), so the join and any lobby admission happen before anyone is talking. There is nothing to launch.

The Calendar page has Day, Week, Month and Year views, filters, and keyboard shortcuts (←/→ to move, T for today, D/W/M/Y to switch view). The view and date live in the URL, so a calendar link can be shared.

Before a call you can attach material to that one meeting: paste a note, or upload documents. Both are stored as extracted plain text and fed into the live prompts, so replies, probes, risks and gaps can draw on your own material.

Both Start meeting and Skip open the same consent dialog, “Save this meeting's context?”, with a tick-box on by default. Where you consent, documents from your earlier meetings with the same client are pulled into later meetings too, newest first.

Seven discovery playbooks ship with the product, read-only:

They are substantial: the Website playbook alone carries 40 checklist items and 30 reference questions. Three parts reach the live AI: the discovery checklist feeds the gaps detector, the top questions feed the next-question generator, and the risk flags feed the risk detector. Items are filtered against your stated role first, so what is irrelevant to your seat drops away.

Your workspace keeps cards with a category, trigger phrases and one approved answer. A new organization is seeded with 31 default cards across all ten categories; solo accounts get their own private library.

Detection runs in two stages: a free keyword match against your trigger phrases first, then an AI classifier reading recent transcript when no phrase hits. When one fires, the panel shows an orange card with the matched phrase, a confidence percentage, your approved answer, and, streaming in below, a version tailored to this client.

When a meeting completes the report generates itself; there is no button to press. It usually takes 10–30 seconds, and the owner is emailed the whole thing inline.

The report has eight tabs:

If a call continues an earlier one, the report adds Related past meetings and Carried over from last meeting, checking each past commitment against the new transcript and marking it done, in progress, not done or unclear. A link the model itself grades low-confidence is discarded, so an uncertain match cannot write another customer's history into your report.

Every way a meeting starts asks the same four things: what type it is, the client it is with, the project it belongs to, and what it is for. A meeting can name a client and a project together, and neither is required. A Procurement meeting names a vendor where the others name a client. A name that does not exist yet is created in the same field, and the title is suggested from the name and the purpose: ABC Company: Product demonstration.

Meetings from before you started filing are not lost. They sit as Uncategorisedand can be filed one at a time, several at once from the Meetings list, or proposed by Brieflee from the call itself. A proposal always says it is one and waits for you to confirm it, so nothing is filed under a heading you did not choose.

What a meeting is filed under is what everything else filters on: the Meetings list, Tasks, Decisions, the analytics and the scope of a question.

Every action item the AI heard, across every meeting, in one list. Tasks are deduplicated by a hash of the meeting and the normalised action text, so re-syncing never duplicates and a task you already ticked stays ticked.

They group into Open, Overdue and Done. Tick them, cycle priority by clicking the chip, or remove them; changes save immediately. Clicking a task opens that meeting's report on the Actions tab and highlights the exact item it came from.

What the AI hears arrives as a suggestion. It waits in Suggested until somebody approves it, one at a time or a whole meeting at once, because “maybe Ahmed can look into this” is not a commitment. A promise made again in a later meeting joins the task already open rather than becoming a second row, and says how many meetings it has come up in.

A daily email is available in Settings, Notifications, with a switch per kind: overdue, due in the next seven days, raised again, waiting for review, and decisions recorded. Every switch is off until you turn it on, and a day with nothing in the kinds you chose sends no email at all.

Ask Brieflee answers from your recorded meetings rather than from one of them. It is in the navigation and on your home page, not buried inside a call: “What did we agree with ABC Company?”, “What is still unresolved on this project?”, “What objections came up most often?”

Narrow it before you ask: everything, chosen meetings, a meeting type, a client, a vendor, a project, a purpose, a participant or a date range. The page says which meetings it searched. Answers carry their sources, and a source card gives the meeting, the date, the speaker and the timestamp; opening a timestamp opens that transcript line, with the recording ready to play from there.

It answers only from meetings you can already open, says plainly when the answer is not in them rather than inventing one, prefers the latest confirmed position when something changed, and explains a contradiction instead of choosing a side. Follow-up questions keep the thread, and a new chat clears it.

What your meetings decided, kept outside the meetings that decided it: the decision, the call it came from, the date and where it stands. Filter it the way you filter everything else, and group it by meeting, client or vendor, project, type or month.

When a later meeting changes its mind, both are kept. The earlier decision is marked as replaced and linked to the one that replaced it, so the chain reads end to end and the current position is never in doubt. Brieflee never connects two decisions on its own: it offers the ones that look related and you choose, becauseuse Provider A for hosting and use Provider A for email read alike and are two decisions.

Fifteen minutes before a meeting that follows an earlier one, Brieflee offers a brief at the top of the app: what was decided, what you promised, what others promised and what nobody took. It is also on the meeting's card in the Calendar and beside your next call on Home, so you can open it sooner.

A meeting follows an earlier one when it is the same repeating meeting on your calendar, or when both are filed under the same project, client or vendor. It reads your own finished meetings, so nothing is offered ahead of a first meeting.

Every finished meeting, including one recorded with no calendar entry, has the same brief at the top of its report: Brief for next call. It opens what that meeting decided and what is still open, with anything left from the earlier meetings it follows.

The brief before a meeting can also come by email, fifteen minutes ahead. It is off until you switch it on in Settings, Notifications, and nothing is sent when there is nothing to say.

Tick off a promise you have kept right there; it is the same task as on the Tasks page. A task Brieflee heard and nobody has approved yet is marked Suggested. Opening a brief costs no credits.

One click on Share mints a public link and copies it. Clicking Share again returns the same link, so a link already sent to a customer is never silently invalidated.

Three CRMs are supported end-to-end: HubSpot, Pipedrive and Zoho CRM. The push lives on the report page and appears only once a summary exists and that CRM is connected.

Every task carries a breadcrumb back to Brieflee: the meeting title, the suggested owner and the summary id. Due dates are read from the wording of the action item (“by tomorrow”, “by next week”, “in 3 days”) and otherwise default to three days out. A push can partially succeed, and you get a per-item report of exactly what was created.

Step-by-step setup for each, with the data shared and how to disconnect: HubSpot, Pipedrive and Zoho CRM.

Meetings cost no credits. Pro includes up to 60 hours of meetings a month and Enterprise up to 120 hours a person, under fair use, and Free includes 3 hours of meetings a month. Credits pay for live coaching, counted from when you switch coaching on in a meeting, which can be halfway through. 3 credits cover up to 60 minutes of coaching, with a 10-minute grace before the next hour is charged, so 70 minutes of coaching still costs 3, and 71 minutes costs 6. Coaching shorter than 5 minutes in a meeting is not charged by itself; such short stretches add up over the calendar month, and every full hour of them uses one coaching hour. The charge is incremental as coaching runs, not up front.

Your plan includes its coaching hours outright (the Coaching hours column below), and those are spent before any credits you have bought, so a pack you buy on the side tops the plan up rather than replacing it.

Annual billing deposits the whole year's coaching hours up front: 24 for Pro, 60 for Enterprise. Credit packs need no subscription and are sold in 4 sizes:

What costs extra

Nothing else debits a credit: reading history, exports and transcript ingest are all covered. Both extras are refunded automatically if generation fails.

There are two independent axes. Your workspace role is owner, admin or member. Separately, a platform-operator flag unlocks the Brieflee console; being a workspace owner does not grant it.

A solo account has no organization and is treated as owner of its own one-person workspace, with a personal credit balance. It can form or join a team later without creating a new account.

Signing in with Microsoft or Google is separate from connecting those accounts. Sign-in requests only openid email profile; no calendar, mail or task permission at all.

Connecting an account under Settings → Integrations is an explicit second step:

From a calendar event Brieflee stores only the event id, the invite's iCalUID, the title, start and end times, the join URL and a cancelled flag. Attendee lists are never stored. “Read from the invitation” in a meeting's agenda reads that one event's description when you press it and keeps only the agenda lines you save; those lines are never sent to AI. Action items reach Google Tasks or Microsoft To Do only when you pick a target list and press Push.

Create a key at Settings → API keys. It needs a name and an end date. The full secret is returned exactly once, at creation; the list endpoint has no field for it, so a lost key is unrecoverable. Create a new one and revoke the old.

Three headers all authenticate a key:

X-API-Key: de_…
Authorization: Bearer de_…
Authorization: de_…

Every path is prefixed with /v1. Start with a smoke test; it returns the account the credential belongs to:

curl https://brieflee.ai/v1/auth/me \
  -H "X-API-Key: de_your_key_here"

Key management

Meetings and transcripts

Asking questions

curl -X POST https://brieflee.ai/v1/meeting-sessions/SESSION_ID/ask \
  -H "X-API-Key: de_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{"question": "What did they say about the timeline?"}'

# => {"answer": "..."}

POST /v1/meeting-sessions/ask-all asks across your recent meetings and returns the same shape. That one costs 1 credit; the single-meeting form does not.

The notetaker

curl -X POST https://brieflee.ai/v1/bot/join \
  -H "X-API-Key: de_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{"meeting_url": "https://teams.microsoft.com/l/meetup-join/..."}'

It creates a meeting session on the fly if you do not pass one. POST /v1/bot/leave pulls the bot out.

Everything else

Errors come back as standard JSON: {"detail": "..."}.

The verbatim Google API Services User Data Policy Limited Use statement is published on the privacy policy.

Recording, video, transcript, summaries and exports are all written to storage you own, encrypted in transit and at rest. It never has to rest on Brieflee’s infrastructure. Set it up in Settings → Storage.

The two delivery paths are different promises. S3, Azure and Google Cloud can issue a signed URL, so the recording is uploaded straight into your storage and is never written to a Brieflee disk. SFTP, OneDrive and Google Drive cannot take an upload signed straight from the recorder, so their uploads are streamed through Brieflee in chunks, in memory, and never written to its disk either. OneDrive recordings then play straight from OneDrive. Which path a given meeting took is recorded against that meeting.

If your storage is unreachable, Brieflee holds the recording and keeps retrying. You choose what happens then: keep a copy with Brieflee, or have nothing kept anywhere else.

Deleting a meeting deletes its transcript and summaries, and calls the bot runtime to destroy the recording it holds, not just the row in Brieflee's database. Deleting everything is self-serve, irreversible, and purges the recordings too. A workspace with storage of its own can choose to keep a deleted meeting's files there (Settings, Storage).

Security-relevant actions are written to an audit trail: sign-ins with method, meeting deletions, recording purges, API-key creation and integration disconnects. The audit writer never records secrets.