I don't suggest. I do.
Heidi's Agentic Mode does the work instead of recommending it. Most AI hands you homework — a draft to finish, a template to fill, a “you should”. Heidi hands you the finished thing: the follow-up drafted and ready to send, the report built from your actual meetings, the research done before you sat down. You delegate, you set how far she can go alone, and you approve what ships.
Waiting for you
Four things done while you slept. None of them sent.
The queue in the morning. Everything done, nothing sent without you.
Not a suggestion. A finished thing.
Hand over the job and Heidi comes back with the actual output — written, built, researched, ready for one click.
Drafted from the call, ready before you've closed the tab. In conversation she can send it from your own mailbox once you say go; working alone, she stops at the draft.
Real documents built from your actual meetings rather than from a blank page — researched, structured and delivered straight into your Files.
Any person or company, fetched from the live web and summed up before the meeting starts.
Created straight into your task list with deadlines attached — and flagged when they slip.
A contact who needs attention gets flagged, and you get pinged — push or Slack — the moment something needs you.
Four levels of autonomy, applied across the board. Nothing irreversible happens because a machine felt confident.
Straight about formats: agentic deliverables land as clean, structured documents in your Files — ready to read, copy or paste anywhere. Client-ready PDF, PPTX, XLSX and DOCX exist too, as meeting exports — a separate feature, not the agent.
Other AI gives you a head start. I give you the finished line. Small difference on a Tuesday, enormous one at six on a Friday.I act. But you decide how far.
Four levels, and the setting applies everywhere rather than task by task — one decision about how much you trust her, not forty. Confirmation is genuinely enforced: when Heidi is told to wait, she waits.
Balanced
One click to shipdefaultShe prepares the whole thing; you glance and approve the external moves. The sweet spot once you've seen a week of her output and stopped reading every line.
Enforced in code, not in the copy: until you confirm, the tool does not run. And working alone, email stops at a draft on every level.
Most people live in Balanced for a fortnight, then quietly move the boring half of their job to Autonomous and never mention it again.Same to-do list. Two completely different evenings.
The work doesn't disappear. It just stops being yours to type out at eleven at night.
The AI gives you a draft. You still write the email.
The quarterly report starts as a blank page on a Sunday.
You research the person in the four minutes before the call.
Follow-ups slip because writing them is nobody's favourite hour.
You're afraid to automate because you can't see what it'll do.
You close the laptop with the busywork still on it.
While you slept, I worked.
Théo delegated three things and shut the laptop. Here's what was waiting when he opened it.
“Follow up with NorthBeam, build the Q3 report, research Marc before my 3pm.”
Written in the tone he set, signed as him.
One click away. Not sent behind his back — autonomous work always stops at the draft.
Pulled from the quarter's calls, structured and readable.
A finished document in his Files — something to read, not an outline to fill in.
Fetched from the live web and summed up.
Sitting in the prep for his 3pm before he'd had coffee.
The Lemaire brief was started too — that one he wanted to finish himself.
of busywork handed off every week, per user
follow-ups left undrafted once it's running
from drafted to done — you approve, it ships
The ~5h and the zero are estimates from beta use, not stopwatch measurements — the one counter the product actually keeps is workflow runs completed.
I did the busywork. Théo did the sleeping. Honestly the better deal for both of us.An AI that acts on its own? Fair to be nervous.
Everyone hesitates here, and they're right to. So here are the objections, answered properly rather than waved away.
“So it just does things without asking me?”
actuallyOnly as far as you allow. Four levels — Careful, Balanced, Autonomous, Full — set once, applied everywhere. Most people live in Balanced until they trust it, then loosen it on the safe, repetitive work.
“It'll email a client at 3am and I'll find out at 9.”
actuallyIt won't, by design. Working alone, Heidi always stops at the draft. She only sends a real email when you're in the conversation with her and you've said go — from your own mailbox, with you watching. The irreversible step is structurally yours.
“Will the emails sound like a robot wrote them?”
actuallyYou set the voice once — tone, length, how much bite — and she signs as you. Every draft follows the register you chose rather than a generic template, and it's right there to tweak before anything ships. No mystery imitation: you stay in charge of how you sound.
“What if it gets something wrong?”
actuallyAt the careful end you see everything before it moves. At the autonomous end you only delegate low-stakes, repetitive work. You never hand over the keys to something you can't take back.
An AI that acts and won't tell you where it stops is a liability. I'd rather bore you with the limits than surprise you with them.Triggered by what happens. Not by a clock.
There's no scheduler where you pick a time and a repeat rule. Work starts because something happened: you delegated a job, a meeting ended, an email came in, a score dropped. The recurring housekeeping — overdue tasks, contacts going quiet — runs on Heidi's own nightly schedule underneath. You describe the outcome, not the cron.
- You delegate — in chat, out loud, or straight after a call
- A meeting ends and the follow-up drafts itself
- An incoming email or a falling score kicks off a workflow
- The nightly housekeeping runs on its own — overdue tasks, quiet contacts
Who is Agentic Mode for?
Anyone whose real job keeps getting interrupted by the admin around it.
Follow-ups drafted the moment the call ends, ready to approve
Prospect research done before the meeting, not during it
The commitments you made turned into tasks before you forget them
Nothing sent to a client without you seeing it first
How is this different from an AI that drafts?
A drafting assistant gives you a starting point, which is still work. An automation tool does exactly what you configured and nothing more. Agentic Mode takes the job, decides how to do it, produces the finished output, and stops at the line you drew.
Written follow-ups, built reports, real research — finished work, on the leash you set.
Delegate the outcome, not the steps. Approve what ships. Nothing irreversible happens without you.
I can write the email, build the report and do the research. I can't decide whether your client should hear from you today. That one stays yours.People who stopped doing their own busywork.
“I delegate the report, go to bed, and wake up to a draft in my Files. It's the first AI that actually finishes the job instead of handing it back to me.”
Consultant · beta
You slept. I worked. I'm not bitter about it.“I started on the careful setting because I didn't trust it. Two weeks later half my admin runs without me and I've stopped reading the drafts line by line.”
Account Executive · beta
Trust me slowly. It's the only sensible way to do this.Which plan includes Agentic Mode?
Agentic Mode starts at Pulse.
Core
49,90 €/mo
39,90 €/mo billed annually
- 10 live meetings per month
- Live Copilot — the basics
- Basic analytics
- Agentic Mode not included
Pulse
89,90 €/mo
71,90 €/mo billed annually
- Unlimited live meetings
- Full Live Copilot + live scoring
- Deep Insights
- Full Agentic Mode — four autonomy levels
Prime
149,90 €/mo
119,90 €/mo billed annually
- API and webhooks
- Unlimited live audio with Heidi
- The full Team view
- Full Agentic Mode
The honest questions.
Updated July 2026
Agentic Mode is the hands.
Twelve features, one Meeting OS. Here's where this one sits.
Stop doing your own busywork.
14 days of the full Prime experience, €0 today. Delegate something tonight and wake up to it done.
Card required · cancel in one click · starts with a ~90-second voice onboarding where Heidi interviews you