AI Brain’s Input Layer

Capture all the company knowledge in online and offline meetings, calls, rooms, screens and ambient office conversation — with quality and consent.

Where it’s said
Captured withMeeting Bot SDKAPIMCPUI ComponentsDesktop SDKAPIMCPUI ComponentsMobile SDKAPIMCPUI ComponentsConnect SDKAPIMCP
Comes back
  • Transcript
  • Who said each line
  • Language of each line
  • Timestamps
  • Participants and the meeting or call
  • Screen context
  • Consent record

Powerful features built in

Notes are only as good as their names.

without voice prints
transcript
  • ?
    Speaker 214:03:22

    If it’s two years we can do twenty-seven.

  • ?
    Speaker 314:03:23

    I’ll put that in the contract tonight.

“Someone will send the terms.”helps nobody

with voice prints
transcript
  • JM
    Joan Martí14:03:22

    If it’s two years we can do twenty-seven.

  • LS
    Laia Soler14:03:23

    I’ll put that in the contract tonight.

“Laia sends the terms tonight.”is a task

Quality engine

You never pick a model. Live scores on real data pick one for each language, inside the SDK.

Your audio

Meetings, calls and rooms, in any of 130+ languages.

Best engine per language

Each language goes to the engine that writes it best today.

10+ speech engines

Our own fine-tuned models, and third-party ones.

Your transcript

Named, timed, in its language.

Behind it, all the time:

600 linguists

Correcting transcripts in real time: the ground truth every engine is scored against.

NetflixElevenLabs

Scores on real data

Every engine, language by language, on real meeting audio.

…and the scores pick the engine.

What you get back

Same four voices. A different transcript.

Four people at one table, as most speech-to-text hears them, and as _æfer does.

Room 3B · 4 voicesillustrative
  1. Speaker 1Pau Ribas

    Legal needs the deep pay signed by Friday.Legal needs the DPA signed by Friday.

    en
  2. Speaker 2Marta Vidal

    I’ll send it to Kenny what an a bay at north wind.I’ll send it to Kenji Watanabe at Northwind.

    en
  3. Speaker 1Laia Soler

    The cord, who posed a contract of email.D’acord, ho poso al contracte avui mateix.

    enca
  4. Speaker 3Joan Martí

    Con dos años bajamos al veinte siete.Con dos años bajamos al veintisiete.

    es
speakers3, one of them twice4, by name
words5 wrongas they were said
languagesCatalan read as Englishen · ca · es
How a company brain fills up

Everything flows into the brain. Except what’s said.

Connectors pour in what’s written. _æfer feeds company brains and agent builders what’s said.

Who builds on _æfer: company brains and agent builders.

connectors Gmail, Outlook Google Calendar Slack Notion, Confluence Google Drive Jira, Linear GitHub HubSpot, Salesforce
_æferwhat’s said

MeetingsCallsRoomsScreens

Company brainsmemory and search
  • Glean
  • Dust
  • Mem
  • Notion AI
  • ChatGPT Enterprise
  • Sentra
agent builders
Sierra support Decagon support HappyRobot operations Harvey legal Cognition engineering Timbal automation

Company names are market examples.

What you get

Five things you ship. One SDK.

Your apps, your components, the transcript and your own data, all in your name.

A desktop app with your name on it.

Mac and Windows. It records meetings and calls from the computer, with no bot in the call, plus what’s on screen.

Apps, components, field names and schemas illustrative.

What you get

Five things, one SDK. All with your name on.

Two apps, the components, the transcript, and your own data pulled from every conversation.

01Desktop appMac · Windows · no bot in the call
02Mobile appin-person meetings
03Meetings components@aefer/ui · drop into your app
MeetingList<MeetingList user={me} />
  • Harbor renewalRoom 3B · in person
  • Pricing syncZoom · Ana, Joan, Laia
  • Call with NorthwindPhone · Marta
propfilterproponSelectproplive
MeetingDetail<MeetingDetail meetingId="mtg_8f2" />

Harbor renewal

Tue 14:03 · 41 min · Room 3B
LSJMMVPR
summary
  • Harbor renews for two years at 27.
  • Laia updates the contract today.
  • Marta sends the DPA to Northwind; signed by Friday.
propsummaryproptabs
Transcript<Transcript meetingId="mtg_8f2" />

LSLaia SolercaD’acord, ho poso al contracte avui mateix.

JMJoan MartíesCon dos años bajamos al veintisiete.

MVMarta VidalenI’ll send the DPA to Northwind tonight.

propfollowproponLineClickpropshowLanguage
RecordButton<RecordButton />
Recording
propsourcepropconsent
SpeakerChip<SpeakerChip id="vp_laia" />
LSLaia Solerby voiceJMJoan Martíby voice
proponRename
04The transcriptnamed · timed · per-line language
MeetingList<MeetingList user={me} />
  • Harbor renewalRoom 3B · in person
  • Pricing syncZoom · Ana, Joan, Laia
  • Call with NorthwindPhone · Marta
propfilterproponSelectproplive
MeetingDetail<MeetingDetail meetingId="mtg_8f2" />

Harbor renewal

Tue 14:03 · 41 min · Room 3B
LSJMMVPR
summary
  • Harbor renews for two years at 27.
  • Laia updates the contract today.
  • Marta sends the DPA to Northwind; signed by Friday.
propsummaryproptabs
Transcript<Transcript meetingId="mtg_8f2" />

LSLaia SolercaD’acord, ho poso al contracte avui mateix.

JMJoan MartíesCon dos años bajamos al veintisiete.

MVMarta VidalenI’ll send the DPA to Northwind tonight.

propfollowproponLineClickpropshowLanguage
RecordButton<RecordButton />
Recording
propsourcepropconsent
SpeakerChip<SpeakerChip id="vp_laia" />
LSLaia Solerby voiceJMJoan Martíby voice
proponRename
05Your data schemadefine it once · extracted from every conversation
your schemaschema.ts
// extracted from every conversation
type Event = {
  event: "decision" | "commitment" | "risk";
  owner: Person;
  due?: Date;
  quote: string;
  timestamp: string;
};
// one row per deal mentioned
type DealStage = {
  account: string;
  stage: "discovery" | "proposal" | "negotiation" | "won";
  amount?: number;
  next_step: string;
  quote: string;
};
// what customers push back on
type Objection = {
  topic: "price" | "security" | "timing";
  customer: string;
  said_by: Person;
  quote: string;
  timestamp: string;
};
// incidents reported out loud
type Incident = {
  system: string;
  severity: "low" | "high";
  reported_by: Person;
  quote: string;
  timestamp: string;
};
extracted · Harbor renewal3 rows
  • decisionWe renew for two years at 27.owner Joan Martí14:03:19
  • commitmentI’ll send the DPA to Northwind tonight.owner Marta Vidaldue today14:03:36
  • riskLegal needs it signed by Friday or the renewal slips.owner Pau Ribasdue Fri14:03:31
  • negotiationTwo years at 27, if they sign this month.account Northwindnext step: send the DPA
  • proposalLet’s send Linden the pilot terms next week.account Lindennext step: pilot terms
  • wonAcme signed this morning.account Acmenext step: kickoff
  • priceTwenty-nine is above what we budgeted.Northwind · Kenji Watanabe14:01:52
  • securityOur legal team needs EU residency.Northwind · Kenji Watanabe14:02:40
  • timingWe can’t start before the board meets.Linden · CFO16:12:05
  • highCheckout has been failing since nine.system paymentsreported by Ana García09:41:10
  • lowThe export button is slow again.system reportsreported by Laia Soler11:20:44
  • lowSearch misses last week’s files.system searchreported by Joan Martí15:05:31

Schemas and fields illustrative. Rows come back over the API and MCP, with the quote and timestamp they came from.

Brand, names, components and schema illustrative.

Speaker identification

Six people, one phone. Every voice named.

No logins in a meeting room, and voices overlap. Voice prints put a name on each one, even offline.

Room 3B · seen from abovein person
Harbor renewal · 41 min0 of 6 named
voice prints, matched on the audio
Speaker identification

One voice, wherever it’s said.

A call, a room, a phone, a pocket recorder. The same voice print names Laia in all of them, so her week joins up.

  • Zoom · Meeting Bot

    “Can we move the launch to May?”

    Mon 10:12
  • Room 3B · laptop

    “Fine, the contract goes out tonight.”

    Tue 14:03
  • Phone call · mobile

    “Northwind signed. Tell legal.”

    Wed 17:45
  • Plaud · Connect

    “Idea: a renewal reminder for every client.”

    Thu 08:30
Laia Solerone voice print · 4 devices
Devices

Pocket recorders, in the same record.

Plaud and Pocket recordings go through the same engines and voice prints as your meetings: more accurate than the devices’ own transcription, and filed right next to them.

recorded on
Plaudsite-visit.m4a · 38 min
Pocketwalk-to-station.m4a · 12 min
run through
_æfer The best engine for the language The same voice prints Connect SDK · API · MCP
Laia’s Thursday, in one record
  1. PocketLaia SolerIdea: a renewal reminder for every client
  2. Pricing syncAna García, Joan Martí, Laia Soler
  3. PlaudSite visit, Harbor’s new officeLaia Soler, Pau Ribas · named by voice
  4. Call with NorthwindMarta Vidal, Laia Soler
Screens

Telepathy for your agent.

The Desktop SDK sees your screen and hears your conversations while you work. You don’t repeat yourself; your agent already knows.

Laia’s screennot seenseen by your agent
harbor-renewal.xlsx
ClientNorthwind
ContactKenji Watanabe
Term2 years
Price / seat27
ClauseEU residency
new email · draft
To: Kenji Watanabe

Hi Kenji, following today’s call

your agent
Finish this email for me

Sure. Which email, to whom, and what should it say?no context

Done: the renewal to Kenji at Northwind, two years at 27 per seat, EU residency clause kept, as in the sheet you just edited.from your screen, 2 min ago

Illustrative. Screen capture can run on the user’s machine, under company and personal rules.

What a company brain sees

Connectors see the tip. The rest was said.

Email, docs and tickets sit above the line. _æfer brings in what’s below it.

Gmail, Outlook Notion, Drive, Confluence Slack Jira, Linear, GitHub HubSpot, Salesforce what’s written ↑what’s said ↓ The client call Monday’s stand-up The 1:1 Design review, Room 3B What was on screen A call from the car Office talk The hallway
above the line · what connectors read
  • Gmail, Outlook
  • Notion, Drive
  • Slack
  • Jira, Linear
  • HubSpot, Salesforce
below the line · what _æfer brings in
  • Client calls
  • Stand-ups and 1:1s
  • Meeting rooms
  • Screens
  • The hallway
_æfer feeds their products
Who builds on _æfer
Company brainsmemory and searchGleanDustMemNotion AIChatGPT EnterpriseSentra
Agent buildersenterprise agentsSierraDecagonHappyRobotHarveyCognitionTimbal
Company names are market examples.Most of what a company knows was never written down.
01

Agents read. But they’re not in the room.

Every decision worth knowing was said before it was written. Most were never written at all. Agents that only read are working from yesterday.

02

Capture everything?

Sales picked one tool, legal another, the room got nothing. One layer captures the work wherever it is said.

MeetingsCallsComputersRoomsChats

Yggy loves privacy.

Yggy is _æfer’s ears. It listens only where your company and each person allow it.

On. Everyone in the room is told.
Off. Yggy sleeps.

Your context. Your rules.

Your agents get the work. Nobody gets watched.

Laia’s Tuesday
Product reviewkept
Call with Acmekept
Pricing doc, with her agentkept
work filter
Call with her mumleft out
Direct message to Joannever read
Booking a dentistleft out
_æfer always_æfer never
Sticks to meetings and workA work filter keeps personal moments out.Reads private chats or direct messages
Tells everyone in the call that notes are onListens in the background without people knowing
Lets each person say no, pause, review or remove what they shareKeeps what someone asked it not to keep
Keeps user rules on, whatever the admin setsLets anyone switch them offNot an admin, not a manager.
Follows your company’s rules for who is recorded, who sees it and where it’s storedThe most specific rule wins.Shares with anyone your rules leave out
Company rulesUser rulesWork filterEnterprise readyEU residency
04

For the teams deploying AI at work.

You’re building agents for a company – yours or a client’s. _æfer gives them what was said there, through one API, so you can spend your time on the agent.

Company brainsAI ConsultanciesInternal AI teams

Stupid simple to install.

  1. 010:00
    Paste

    One prompt into your coding agent.

  2. 020:12
    Wire

    It reads the docs and writes the code.

  3. 033:52
    Approve

    You say yes to the key.

  4. 044:51
    Done

    _æfer is in your product.

HeadlessAPIMCPApp

Context infrastructure for enterprise agents.

Capture the meetings, calls and rooms your rules allow, so agents can act on them. Ship in 5 minutes.