The parts the tour skipped.

Pick the language per meeting, or let it detect one. The list is served from the backend to the picker, so the app can never offer a language the recogniser would reject, and the choice applies the same way whether you send the notetaker by hand or your calendar does. Arabic and Urdu lines lay out right-to-left in the live panel and in the finished report.

• EnglishEnglish
• AfrikaansAfrikaans
• العربيةArabic
• অসমীয়াAssamese
• БеларускаяBelarusian
• БългарскиBulgarian
• বাংলাBengali
• BosanskiBosnian
• CatalàCatalan
• ČeštinaCzech
• DanskDanish
• DeutschGerman
• ΕλληνικάGreek
• EspañolSpanish
• EestiEstonian
• فارسیPersian
• SuomiFinnish
• FrançaisFrench
• ગુજરાતીGujarati
• עבריתHebrew
• HrvatskiCroatian
• MagyarHungarian
• ՀայերենArmenian
• Bahasa IndonesiaIndonesian
• ItalianoItalian
• 日本語Japanese
• ქართულიGeorgian
• ҚазақKazakh
• ಕನ್ನಡKannada
• 한국어Korean
• LietuviųLithuanian
• LatviešuLatvian
• МакедонскиMacedonian
• МонголMongolian
• मराठीMarathi
• Bahasa MelayuMalay
• नेपालीNepali
• NederlandsDutch
• NorskNorwegian
• ਪੰਜਾਬੀPunjabi
• PolskiPolish
• پښتوPashto
• PortuguêsPortuguese
• RomânăRomanian
• РусскийRussian
• SlovenčinaSlovak
• SlovenščinaSlovenian
• СрпскиSerbian
• SvenskaSwedish
• தமிழ்Tamil
• తెలుగుTelugu
• ไทยThai
• TagalogTagalog
• TürkçeTurkish
• УкраїнськаUkrainian
• اردوUrdu
• Tiếng ViệtVietnamese
• 中文Chinese

What it does not doAuto-detect covers 10 languages and is deliberately not the default: on short meeting turns it has turned English into Devanagari. Naming your language is materially more accurate.

A solution architect and an account executive on the same call get materially different suggestions, because the role rewrites what the model is asked for, not just how the answer is dressed. Add free-text context about your seat and it goes in too.

• Account executive
• Pre-sales
• Sales engineer
• Solution architect
• Consultant
• Project manager
• Product manager
• Developer
• Tech lead
• CTO
• Marketing manager

An internal standup or a support call returns nothing at all, and that is enforced in code rather than asked for in a prompt. When it is a deal call, it judges what is missing against the stage you are actually at, so it will not flag an unnamed budget holder on a first intro call and train you to ignore the panel by minute four.

Meetings export as Markdown, plain text, JSON, or a self-contained discovery summary laid out for print. The client-facing document is assembled from an explicit list of what may appear, so the risk analysis, the objection handling and the coaching notes about the person you are sending it to are structurally incapable of ending up in it; it is not a redaction step that can be forgotten.

What it does not doThere is no server-side PDF renderer. The print layout is HTML your browser turns into a PDF.

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 a line saying what it is for. A meeting can name a client and a project together, and neither is required; a procurement call names a vendor where the others name a client. A name that does not exist yet is created in the same field without leaving the form, and the title is suggested from the name and the purpose, so a call arrives as “ABC Company: Product demonstration” rather than “Meeting”. Meetings you already have are not lost or blocked: they sit as Uncategorised, can be filed in bulk, and Brieflee will propose a type and a purpose for them from the call itself. A proposal is always shown as one and waits for a person.

What it does not doThis is context for meetings, not a sales or procurement system: a client, a vendor or a project is a name and the meetings it appears in.

Ask Brieflee sits in the navigation and on the home page rather than inside a meeting, because a question is rarely about one call: what did we agree with this client, what is still unresolved on this project, what changed since last month. Narrow it to a type, a client, a vendor, a project, a purpose, a person or a date range, and the answer arrives with its sources: the meeting, the date, the speaker and the timestamp. Open a timestamp and the transcript opens at the moment it was said, with the recording ready to play from there. It answers only from meetings you can already open, says so plainly when the answer is not in them, prefers the latest confirmed position and explains a contradiction rather than choosing a side. Follow-up questions keep the thread.

What it does not doAsking about one chosen meeting is free. Any wider question is priced as an AI question, and a question with nothing to search costs nothing.

Tasks gathers what was promised across every meeting, grouped by the call it came from, with an owner, a due date, a priority and a link to the line it was said on. What the AI hears arrives as a suggestion and stays there until somebody confirms it, because “maybe Ahmed can look into this” is not a commitment. A promise made again in a later meeting joins the one already open instead of becoming a second row, and says how many meetings it has come up in. Filter by owner, status, due date or how the meeting is filed, push a selection out to wherever your team already tracks work, and switch on a daily email covering only the kinds you choose.

What it does not doA repeat is joined only when the wording is close enough to be sure. It would rather leave two rows you can merge than merge two promises you made.

Every decision a meeting reaches is recorded on its own page: what was decided, in which call, on what date, and where it stands now. When a later meeting changes its mind, the earlier decision is not deleted. It 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. Nothing connects two decisions on its own: Brieflee offers the candidates and a person chooses, because “use Provider A for hosting” and “use Provider A for email” read alike and are two decisions.

What it does not doDecisions are read from the meeting reports, so a meeting with no report has none to record.

Upload mp3, m4a, wav, mp4, webm, ogg or flac up to 100 MB, which is about two hours as mp3, or a pdf, docx, txt or markdown document, or paste a transcript or your own notes straight in. It is transcribed and summarised in the background, so the page does not have to stay open, and the result is filed, searchable and exportable like every other meeting.

What it does not doSpeakers in an uploaded recording come back as Speaker 1 and Speaker 2, and are renamed on the transcript.

The dashboard opens on what is next: your upcoming calls with the notetaker toggle, what was promised since you last spoke to these people and whether it happened, deal momentum, the verbatim buying signals worth re-reading, and where your hours are going. The library is searchable, filterable and refreshes on its own, so a call that has just ended is there without a reload. Analytics cover meeting volume and cadence, recorded time, signals, follow-up completion, sentiment, buying signals and recurring objections over any date range.

What it does not doThe per-teammate breakdown needs a workspace owner or admin, and team activity only appears once a workspace has more than one member.

One-click presets for Amazon S3, Cloudflare R2, Wasabi, Backblaze B2, DigitalOcean Spaces and MinIO, plus Azure Blob, Google Cloud Storage, any SFTP server, and OneDrive or Google Drive, which you connect by signing in rather than with keys. Set a folder prefix, server-side encryption and your own KMS key. Saving does not just store the credentials: it writes a canary object to the destination, reads it back and deletes it, then tells you which of the three failed. You also choose what happens if your storage is unreachable: keep the recording with us, or refuse to write it anywhere else.

Saved credentials are write-only. They come back masked and stay that way, so editing a bucket prefix cannot wipe your access key.

A recording made on a phone is handed a signed URL and uploaded straight into your bucket; it is never written to our disk. A recording the notetaker made is streamed to your storage in chunks through memory, never persisted on the way, and the recorder’s own copy is deleted afterwards. Which of the two happened is stored against the meeting, so the answer to the question in the questionnaire is a lookup rather than a paragraph of hedging.

What it does not doSFTP, OneDrive and Google Drive are always the relayed path: none of them can take an upload signed straight from the recorder.

The durable copy of an uploaded recording is encrypted on our filesystem, so a stray volume snapshot or backup is not a playable meeting.

What it does not doScope is the stored segments; coverage of other stored objects is not complete. The exact scope is something we set out in a security review rather than on a public page.

Each workspace gets its own entity ID and callback URL, so an assertion minted for one tenant cannot be replayed into another. SCIM runs at the standard endpoint, which means deprovisioning is your identity provider’s job rather than a checklist item someone forgets on a Friday.

What it does not doNot self-serve yet. There is no admin screen for the identity-provider fields, so we set them up with you.