Mimi AI
The AI assistant inside every customer conversation.
Mimi reads the thread so you don't have to, drafts the reply you were about to write, and works only where your permissions already let you work.
What Mimi does
Built for the work your team repeats every day.
Two things are ready now. The rest is on the way, and each is marked honestly below.
Three weeks of back-and-forth, understood in ten seconds.
Ask Mimi to summarise and she answers the three questions that matter when you pick up a conversation: what the customer wants, what your team has already done, and what they are waiting for. Handovers, holidays and escalations stop costing a full re-read.
A first draft in context. You decide what goes out.
Draft with Mimi writes from the conversation and the customer's history. Anything she cannot know, like a date or an order number, is marked [[like this]] for you to fill in, and links she does not recognise are called out. Insert it, regenerate it or discard it; it only reaches the reply box when you choose.
The right inbox and the right person, first time.
Today MimiDesk routes with rules you set. Mimi will suggest where a new conversation belongs and who should own it, and, with your permission, apply it.
The CRM notices what the conversation said.
Contacts and companies already build themselves from real mail. Mimi will propose the updates a conversation implies, like a new phone number, a renewal date or a change of role.
The questions you have answered a hundred times.
For the routine questions you approve, Mimi will answer and close the conversation herself, within an allowance on your plan, and report every one.
Your data
Exactly what Mimi sees, and what is never sent.
The same list sits on your admin settings page, before anyone switches her on.
Sent, when someone asks
- Only when a person asks: nothing is sent on a schedule
- Only the one conversation you are looking at
- Recent messages, the contact's name and timestamps
- For a summary, a few internal notes for context; for a draft, none
Never sent
- Attachments
- Email addresses and card numbers: replaced before anything leaves
- Conversations you are not allowed to open
- Anything at all while Mimi is switched off
Every request is written to the audit log: who asked, which conversation, how much was sent and where. Prompts and answers are not stored.
How far she goes
You set the authority. Mimi works within it.
Autonomy is a permission, never the default. Each workspace chooses how far Mimi may go, can change it any time, and sees every step she takes.
01
Suggest
Summaries and drafts. A person reads, edits and sends.
Available
02
Prepare
Tags, tasks and replies set up for a person to confirm.
On the way
03
Act
Routine work your admins switch on, done and recorded.
On the way
04
Resolve
Closing the conversations you allow, within your plan's allowance.
On the way
Human control
Three gates before a single word reaches a model.
Every part of MimiDesk works with Mimi switched off. She adds to the workflow; she never gates it.
How AI meets your data- 1Your plan includes MimiShe is an add-on on Scale and Enterprise. Without it, no part of MimiDesk calls a model.
- 2An admin switches her onOff by default for every workspace. The switch says exactly what is sent and where before anyone confirms.
- 3A person asksEvery summary or draft starts with someone pressing a button in a conversation they can already open.
Principles
The rules Mimi lives by.
- 01Never annoying.No popups, no mascot, no forced AI moments. Easy to reach, quiet when she has nothing useful to add.
- 02Honest when unsure.Gaps are marked, not guessed. Confidence is earned by accuracy, never performed.
- 03No fake intelligence.If a screen says Mimi did something, she did. Ordinary saving and syncing is never dressed up as AI.
- 04Not a permission bypass.Mimi sees what the person asking can see, and nothing more.
Mimi is an add-on on Scale and Enterprise, priced per agent seat on the same billing term as your plan.
See Mimi pricingQuestions
What teams ask about Mimi.
What is Mimi?
Mimi is the AI assistant built into MimiDesk. She works inside your customer conversations rather than in a separate chat window: today she summarises long conversations and drafts replies, and more of her work is on the way, always within the permissions your workspace sets.
Does Mimi send emails to customers by herself?
Not today. Summaries and drafts are suggestions: a draft only reaches the reply box when a person inserts it, and a person sends it. As Mimi takes on routine work, each step will be something your admins switch on deliberately, and every action will be recorded.
What customer data does Mimi send to the AI model?
Only when someone asks, and only from the one conversation they are looking at: recent messages, the contact's name and timestamps. Attachments are never sent, and email addresses and card numbers are replaced before anything leaves. The admin settings page lists exactly what is sent and where.
Can Mimi read conversations I don't have access to?
No. Mimi reads through the permissions of the person asking, so she cannot see a conversation that person cannot open. AI is not a way around your access rules.
Is Mimi switched on by default?
No. Mimi is off for every workspace until an admin turns her on, and turning her on asks for confirmation. Switching her off stops anything further being sent immediately.
Is there a record of what Mimi did?
Yes. Every request is written to the audit log: who asked, in which conversation, how much was sent and where. Requests Mimi declined are logged too. The prompts and the model's output are not stored.
How much does Mimi cost?
Mimi is an add-on for the Scale and Enterprise plans, priced per agent seat and billed on the same term as your plan. Every part of MimiDesk works without her.
Meet Mimiin your own inbox.
Join the waitlist. We'll write when it's your turn — nothing else.
Rather talk first? Contact us