Zeno
Voice agent for enterprise meetings

Ask your systems a question in the meeting. Hear the answer, with its source, before the meeting moves on.

Zeno joins your Zoom, Teams or Google Meet call as a participant. Someone says “Zeno, what were closed-won sales last month?” and it answers out loud from live Salesforce data, names the source and the definition it used, and posts the exact figure to the chat. It is connected to your systems of record, not to a transcript.

ZoomMicrosoft TeamsGoogle MeetWebexSlack huddles · chatIn-country deployment
Zeno
Say my name to ask a question
Listening
what were closed-won sales last month
Weekly sales review · ZoomZeno (AI assistant) in the room
Layla
Zeno — Checking that now. · 0.4s
Zeno
Omar
Zeno
chat · Exact: August 2026 4,213,455 AED (17 deals) · July 2026 4,478,120 AED (19 deals) · Definition: Sales = closed-won by close date, UAE + KSA · Source: Salesforce

Live simulation of one exchange. The dark panel is what the room sees as Zeno’s camera: the question it heard, then the answer with its source.

Not a note-taker

Note-takers summarise what was said, afterwards. Zeno changes what happens during the meeting: the figure that would have been “I’ll check and come back to you” is spoken in the room, from the system of record, while the decision is still open.

Never confidently wrong

Every spoken number must trace to a query result; the software enforces it, not a prompt. Ambiguous term? It asks one question. No matching data? It says “no closed deals in that period”, which is a different answer from “I don’t know”.

Decisions reach your systems

“Zeno, assign those five tickets to Ahmed.” With write actions enabled it confirms aloud, posts a card to the chat, executes on your say-so, reports exactly what succeeded, and can undo. Pilots start read-only.

How it works

Four steps. The first word arrives in under half a second.

  1. 1
    Invite it

    Paste the meeting link in the console, or toggle it on for a scheduled meeting. Zeno joins as a visible participant named “Zeno (AI assistant)” and announces itself in the chat. Nothing to install for anyone.

  2. 2
    Ask, mid-conversation

    Say its name anywhere in a sentence. It knows the difference between “ask Zeno later” and “Zeno, what’s the pipeline?”. Follow-ups within fifteen seconds need no name. “Zeno, stop” is instant.

  3. 3
    See what it heard

    Within half a second it acknowledges out loud and its camera shows the question as it understood it. The room catches a misheard question before the answer, not after.

  4. 4
    Hear the answer, with the source

    “From Salesforce, closed-won for August: 4.2 million dirhams.” Exact figures, period, currency and the definition applied go to the chat. If a query takes longer than eight seconds the answer goes to the chat instead, with a spoken heads-up, so the meeting is never held hostage.

Time from the end of the question
design budget · measured per interaction in the console
Question ends
0 ms
Intent classified
~300 ms
Acknowledges out loud
< 500 ms
Known question answered (pre-fetched)
~1 s
Live query answered
2 – 6 s
Longer than 8 s? Answer goes to chat
8 s

Silence is what makes a voice agent feel broken, not duration. The first sound arrives before the data does; the room hears the source name before the number.

Security and data flow

Your security team’s questions, answered before they ask them.

Where processing happens
Meeting capture and live speech-to-text run in the meeting-bot service in the region chosen for your deployment (US, EU or Tokyo). Video is never captured. Reasoning runs on Anthropic’s Claude API under commercial terms. Voice runs on ElevenLabs. Your data sources are queried through a read-only service account your IT creates, from a connector service you can host in-country.
What is retained
No video. The meeting-bot service keeps the transcript for a configurable window (default 7 days, or zero retention). The transcript and every question/answer chain — who asked, the resolved query, what the source returned, what was spoken, latency per hop — are kept for a configurable period (default 30 days) in your database, then purged.
Permission model
v1 answers only from a dataset your team pre-approves, per client. External participants are detected on join; by default the agent limits itself to public data or stays silent, and says so in the chat. Per-asker and lowest-rights-in-room models arrive with SSO.
Audit
Every spoken figure is reconstructible. Anyone in the room can say “that’s wrong” and the answer is flagged for review with its full chain.
Write actions
Off by default. When enabled: spoken confirmation plus a chat card, execution only on an explicit yes from the requester, an exact success/failure report, and an undo window. Silence cancels.
Full security overview
YOUR MEETINGZoom / Teams / Meetthe bot joins as a participantMEETING BOT SERVICECapture + speech-to-textUS, EU or Tokyo · no video capturedaudioYOUR TENANT (cloud or in-country)Core agentwake word · routing · answer guard · audit log · permission modeltranscript, speaker namesvoice + visual cardClaude APIreasoning · no training on dataElevenLabstext → speech onlyAudit databasequestions, results, latencyquestion + tool resultsanswer textYOUR SYSTEMS (read-only service account)SalesforceMCPJiraMCPWarehouseMCPYour agentsMCP
Why this, and not a meeting assistant

The difference is what it is connected to.

01
Live enterprise data, not meeting transcripts

The answer comes from Salesforce, Jira, Dynamics, Snowflake or your internal agent layer at the moment you ask. The twenty or thirty questions your team asks every week are pre-fetched, so they return in about a second.

02
Works across Zoom, Teams, Meet and Webex

One agent, every platform your people actually use. It joins as a participant; there is nothing to install on anyone’s laptop and nothing for IT to roll out.

03
In-country deployment available

The data-source connectors and the audit database run where you need them. Meeting media can stay in-region, or fully self-hosted for clients with residency requirements.

04
Built on your definitions

“Sales” means what your CFO says it means: which date field, which entities, which currency. Definitions are confirmed by a named person and read aloud when they matter.

05
Arabic on the roadmap

English today. Arabic and English code-switching speech recognition is the next language step, prioritised with the first client that needs it.

06
Delivered as an engagement

We connect your systems, interview your team for definitions, validate every catalogued question with you, and run a dry run before it goes near executives.

Use cases

Decisions reach your systems before the meeting ends.

What reliably fails in meetings is not knowing the number. It is that the decision never makes it into the system. Read-only answers get the pilot in the door; write actions are why it stays.

Operations review
write action

“Assign the unassigned P1 tickets to Ahmed.” Two Ahmeds? It asks which. Confirms aloud, executes in Jira, reports that OPS-13 failed on permissions, offers undo.

Sales pipeline call
write action

“Move Gulf Energy to negotiation and set the close date to the 30th.” Spoken confirmation, chat card, done in Salesforce while the deal is still being discussed.

Executive committee
read

“What were closed-won sales last month, and the month before?” Two answers, source named, exact figures in chat, “as of this morning” when the warehouse is nightly.

Board pre-read
read

“How many open audit findings are older than 90 days?” The room hears the count; the full list lands in the chat.

Finance close
roadmap

“Book the accrual we just agreed.” Not in v1. Writes to systems of record with approval chains are on the roadmap, behind the same confirm-and-undo path.

Anything ambiguous
read

“What’s our headcount?” With no confirmed definition it asks: “Active employees at month end, UAE and KSA?” Exactly one question, never a guess.

Run a pilot on one recurring meeting.

Four to six weeks. Your systems, your definitions, your people. We do the setup with your IT team, validate every answer with your business owners, and measure time-to-answer and accuracy in the meetings that matter.

No signup, no free tier, no self-serve. Every deployment is a conversation.