Product of F51 Labs · Montréal, Québec · Pilot stage

MIKE
A local, verifiable AI voice agent for Québec businesses.

The core conversational AI runs locally. MIKE answers, understands, performs controlled actions and keeps verifiable evidence of critical actions.

Start with a 30-minute conversation. Response within two business days; scope and fees agreed in writing.

MIKE / owner consolepilot interface
MIKE owner console, captured on 23 August 2026
Real interface capture · 23 August 2026. Shows the client cockpit; it does not prove a current customer deployment.
FR-CA + EN Local AI Controlled actions Cryptographic audit Supervised pilots

Pilot stage. The interface shown here is a dated capture, not a live customer deployment.

52
Catalogued skills
13
Configured MCP servers
FR/EN
Languages
5,769
Offline tests
35B
Local MoE
0
Paying customers yet

That last number is not a typo.

01A call, with a human decision

See where MIKE stops and asks.

Replay a bounded example of an incoming call.

call · inbound, Tuesday booking —

This illustration replays pre-computed call steps. It is not a live call or a receipt from a customer deployment.

STATUS AND LIMITS

A supervised pilot, with work still to validate.

Voice latency, Québec French over the phone and simultaneous calls still require validation. The dated measurements and their limitations are public.

Read the measured evidence ↗
06The pilot

One workflow. Four weeks. Bounded before it starts.

A pilot is a fixed scope with a fixed end date, agreed in writing before anyone starts.

Four weeks.
One workflow.

Fixed fee, quoted in writing after a 30-minute scoping call.

  • Week 1 — we bound the workflow and connect only what it needs
  • Weeks 2–3 — shadow mode on your real work
  • Week 4 — measured report with the real absorption rate
  • You keep the full journal, exported
  • Not included — production deployment or new integrations

What happens if it doesn't work

If the measured absorption comes in under 30 %, we say so in writing and do not propose a deployment.

This is a small independent practice, not a platform with a sales quota.

After a successful pilot

Deployment is quoted against the scope the pilot actually measured.

Start with one workflow
07Straight answers

The questions you'd ask on the call.

What does a pilot cost?

A fixed fee, quoted in writing after a 30-minute call.

You get the figure before you commit to anything.

What if it sends something wrong to a client?

Shadow mode means MIKE proposes and a human sends.

The journal tells you exactly which action and who approved it.

Why not just use ChatGPT or Copilot?

MIKE's proposed distinction is a local runtime with evidence on your disk.

If your bottleneck is writing, you don't need MIKE.

What happens to my system if F51 disappears?

MIKE runs on your hardware, in open formats — SQLite, YAML, NDJSON.

What you would lose is support and new skills, not the operation.

Do I need a server room or a data scientist?

A single workstation with a consumer GPU handles a small business workload.

Who has done this already?

Nobody, as a paying customer.

Being among the first is a real risk, and the pilot is structured so that it is a small one.

Can it work in French?

It replies in whatever language the customer wrote in.

08Request a pilot

Give MIKE one workflow.

One real workflow, measured in shadow mode. Tell us what slows your team down; we will scope the pilot with you.

Reply within two business days.

Send directly to the private F51 laboratory inbox. You will see a receipt once your request is saved. We use your contact details only to discuss this pilot.

A sentence or two is plenty.
Voice architecture and measured limitations
◆MIKE Voice

What happens between the ring and the answer.

MIKE answers the phone in French and English, understands what the caller wants, writes into MIKE’s own agenda and keeps a receipt of every action it takes.

Caller — dials the business number
Telephone carrier — the carrier carries the audio to you. This one link is never local.
MIKE — on your machine, on your GPU
├─ Voice activity detection
├─ Turn detection
├─ Speech recognition — Whisper, local
├─ Language model — 35B MoE, local
├─ Actions — calendar, messages, reminders
├─ Promise guard — no claim without a receipt
└─ Voice synthesis — Kyutai, local
Caller — hears an answer, and gets a confirmed appointment

Working today

FR-CA / ENFrench and English, in the same product.
Continuous listeningThe microphone stays open while MIKE is speaking.
Barge-inIf the caller talks over it, MIKE stops and listens.
Agenda actionsBooking, moving and cancelling are executed in MIKE’s own agenda, with a receipt.

In validation — said plainly

Cascade, not native duplexThe pipeline uses continuous listening and interruption.
LatencyThe measured voice path is still above target. See the measured latency →
Concurrent callsOne call at a time today.
Québécois French on callsBelow target on a Québécois telephone corpus.

Real-time voice architecture under validation, with continuous listening and speech interruption.

Explore a cost estimate
01The arithmetic

First, what the manual version already costs you.

Move the three sliders to your real situation.

8 h
2
45 $

Loaded cost means salary plus payroll taxes, benefits and overhead.

Cost per year, doing it by hand
37 440 $
22 464 $
Portion a mission could absorb
499 h
Hours returned to your team per year
Stated assumption: 60% absorption of the repetitive portion.
An example of a working day
02The difference

A chatbot answers. An operator finishes.

You don't hire people to answer questions. You hire them to close loops.

It plans, acts, verifies, remembers and follows up.

07:00
Briefing sent
09:14
Call answered
09:14
Understood in French
09:15
Slots checked
09:15
You approve
09:16
Booking written, receipt issued
12:00
Silence audited
15:00
Follow-ups out
17:30
Day summarised

Shaded = the only moment that needs a human.

Ask MIKE about the product
—Talk to the agent

Ask it what it is. It answers as itself.

A live MIKE instance answers questions about the product below.

mike · product Q&A not checked

What it answers: what MIKE does, how a mission runs, how a pilot is priced.

This demo sends your question through Cloudflare to F51 Labs' machine. Replies are generated and can be wrong. Do not enter private information.

Explore a workflow
03Mission mapper

Describe your bottleneck. Get the real plan back.

Ordered steps, the tool each one needs, what gets checked, where you sign off.

Runs entirely in your browser.

Inspect the demonstration audit trail
04The evidence layer

Don't trust the log. Break it.

Every row carries a real SHA-256 hash computed in your browser.

Not verified yet.
Time Tool Action — editable Result Check Hash (prev → this)
Data location and privacy
05Local-first

Your client history shouldn't become someone else's training data.

Here is the honest split.

Stays on your machine

35B MoEThe model that reads your data runs on your GPU.
Speech in, speech outSpeech recognition and voice synthesis run on your hardware.
SQLite · Mem0Persistent memory of clients, prices and decisions.
LightRAGYour documents, indexed and searched locally.
JournalThe audit trail, on your disk, exportable at any time.

Leaves, because you connected it

EmailEmail leaves the machine only when a mail route is configured. A first-party server and domain are the intended setup; sending and receiving still need proof in each pilot.
Twilio · TelegramMessages transit through the carrier you chose.
Any connector you enableMIKE's own agenda is local. An external calendar, email service or client system is a separate connector, scoped before a pilot starts.
Not claimedThat zero data ever leaves.

Designed with Québec's Law 25 in mind.