Privacy Policy

Effective August 26, 2026 · Last updated August 26, 2026

Mainroom puts AI agents into live meetings, so this policy is mostly about one thing: what happens to what gets said in those meetings. We've written it in plain language and kept it specific to what the product actually does.

The short version

An agent only joins a meeting when someone invites it, and a person in the meeting has to admit it from the lobby.
We don't store recordings of meeting audio or video. We store the text transcript, and the notes and reports built from it.
Private messages sent directly to an agent in meeting chat are never stored.
We don't sell your data, and we don't use your meeting content to train AI models.
Meeting content is processed by AI providers (listed below) to make the agent work.
You can ask us to delete anything, at hello@mainroom.ai.

Who this covers

This policy applies to mainroom.ai, app.mainroom.ai, the Mainroom MCP server at mcp.mainroom.ai, and the agents that join meetings on your behalf. Mainroom is operated by [LEGAL ENTITY NAME], [JURISDICTION].

Two groups of people show up in this policy: customers (people who create an account and build agents) and meeting participants (everyone else in a call an agent joins). Where the rules differ, we say so.

What we collect

Meeting data

When an agent is in a meeting, we process and store:

Transcript text — what was said, with the speaker names the meeting platform reports.
A running summary of the meeting, generated by AI so the agent can recall earlier discussion.
Public meeting chat messages that the agent can read, and messages the agent posts.
Shared-screen images — only if the agent's owner enabled screen vision. We keep only the most recent frame while the meeting is running, replaced as new ones arrive.
What the agent itself said, showed, and generated — its lines, slides, and images.

We do not store meeting audio or video recordings. Audio is converted to text as the meeting happens and the audio is not retained by us. Where an agent has screen vision turned on, we keep a single current screenshot rather than a video stream.

Contact and invitation data

The email address and display name of whoever invited the agent (from the calendar invite or forwarded email).
The attendee email addresses listed on the calendar invite, when the agent's owner has chosen to send the follow-up email to everyone on the invite.
The meeting link and the meeting's scheduled time.

Account and agent data

Your email address, used to sign in.
Agent configuration: names, personas, voice, instructions, meeting manners, rubrics, and settings.
Credentials for tools you connect (for example an issue tracker or CRM). These are stored on our servers and are never sent to the agent's in-meeting page, which only ever sees the tool's label.
If you set up a "connected browser" for demos behind a login, the browser session established during that setup is stored with our browser-infrastructure provider.

Notes an agent keeps about people

If an agent has memory enabled, after each session it writes short private notes about the people it worked with — what they practiced, strengths, focus areas — so the next session can pick up where the last one left off. These notes are visible to the agent's owner in the follow-up email and are used to inform later sessions. Memory can be turned off per agent, and is off for the public demo agent and for agents created through our MCP server.

If you are the subject of those notes — for example someone an agent coached — you can ask us for a copy of what an agent has recorded about you, or ask for it to be deleted, at hello@mainroom.ai. You do not need a Mainroom account to make that request.

Technical data

Standard server logs (IP address, timestamps, error diagnostics) from our hosting and edge providers, used to operate and secure the service.

What we don't collect

Private whispers. If you direct-message an agent in meeting chat, that message and the agent's reply are relayed to the agent in the moment and are deliberately excluded from stored transcripts, summaries, and post-call reports.
Recordings. As above — no stored audio or video.
Payment card details. We don't currently process payments.
Special-category data. We don't ask for it. If it's spoken in a meeting, it lands in the transcript like anything else, so please invite agents accordingly.

How we use it

To make the agent work during the meeting — understand what's said, respond, present, research, and post to chat.
To produce the post-call email the agent's owner configured (a summary, coaching notes, a decision list, a risk register, and so on).
To let an agent remember people across sessions, when memory is enabled.
To operate, secure, debug, and improve the service.
To contact you about your account or a session you were part of.

AI processing and who else sees the data

Mainroom is built on third-party infrastructure, and running an agent means meeting content passes through some of it. These are our subprocessors:

Provider	What it handles
Recall.ai	Joins the meeting, provides the agent's camera and microphone presence, and streams captions, chat events, and (when vision is on) screen images
OpenAI	Speech-to-text, the language model that decides what the agent says and does, and text-to-speech
ElevenLabs	Text-to-speech for certain voices
Google (Gemini)	Generating images the agent shows on screen
Supabase	Database, authentication, and server functions
Cloudflare	Website and API hosting, and email routing for agent addresses
Resend	Sending post-call emails and agent replies
Browserbase	The live browser an agent drives during a demo
Exa	Built-in web search when an agent is asked to look something up
Wikimedia / Openverse	Finding real photos when an agent is asked to show one (we send the search terms only)

If you connect your own tools to an agent (an issue tracker, CRM, analytics, or any other connector), that provider receives whatever the agent sends it when carrying out a task you asked for. Those providers are governed by your agreement with them, not by this policy.

Model training

We do not use your meeting content, transcripts, or agent configurations to train AI models, and we do not sell or rent your data. We use our AI providers under terms intended to prevent your content being used to train their models. We can't control what a tool you connect does with data you direct an agent to send it — check that provider's terms.

What an agent can do in a meeting

A Mainroom agent is a participant, not a passive notetaker, and that's worth understanding before you invite one. Depending on how its owner configured it, an agent can:

Speak aloud in the meeting, so what it says is heard by everyone present.
Put slides and images on the shared screen, including material it writes or generates during the call.
Run a live browser demo on the shared screen.
Read and post messages in the meeting chat.
Search the web and use tools its owner connected — which can mean creating or updating real records in those systems (an issue, a CRM note) as a result of what's said in the meeting.

Owners choose which of these abilities an agent has, and anyone in the meeting can tell it to go on mute. Like any AI system, an agent can be wrong: it can misunderstand what was said, state something inaccurate, or take an action you didn't intend. Treat its output as a starting point to check rather than a system of record, and give agents access only to tools where that's an acceptable risk.

Agents that arrive without an account

Two paths let an agent join a meeting without anyone signing up: our public demo agent (demo@mainroom.ai), which anyone may invite, and our MCP server, which lets an outside AI assistant give itself a temporary meeting presence. Because these are open, they're deliberately fenced:

Whoever dispatches an agent must already have the meeting link.
A person in the meeting still has to admit it from the lobby — it cannot let itself in.
Dispatches are rate-limited per meeting and per requester.
These paths can only summon the public demo agent or a temporary body. They can never reach an agent belonging to a customer account, or that account's data.
Temporary bodies keep no memory, send no email afterwards, and are deleted within 7 days.

If an agent appears in a meeting and you don't know why, don't admit it — or remove it as you would any participant — and tell us at hello@mainroom.ai.

Consent, notice, and recording laws

If you invite an agent to a meeting, you're responsible for having the right to do so. That includes telling the other participants that an AI participant is present and processing what's said, and getting consent where the law or your organization requires it — some places require every party's consent.

Mainroom is built so this is visible rather than hidden:

The agent appears in the participant list under its own name, with a camera tile.
A human has to admit it from the meeting lobby — it can't let itself in.
Agents can be configured to introduce themselves in chat or out loud when they join.
Anyone in the meeting can tell an agent to go on mute, and it stops speaking and presenting.

If you're a participant and you'd rather an agent weren't there, ask the organizer to remove it, or remove it from the meeting the way you would any other participant.

Who receives the post-call email

Each agent's owner chooses where its follow-up email goes: to the person who invited it (the default), to a fixed address such as a team alias, or to everyone on the calendar invite. The email contains what the agent was configured to report — which may include summaries of what participants said and, for coaching agents, observations about them. Choose this setting with the people in the room in mind.

How long we keep things

Meeting transcripts, summaries, reports, and agent memory — kept while your account is active, so agents can reference past sessions. Deleted on request.
Screen images — only the latest frame during a live meeting; superseded as new frames arrive and removed with the session record.
Agents created through our MCP server (an outside AI giving itself a temporary meeting presence) — these keep no memory, send no email, and are deleted automatically within 7 days.
Account and agent configuration — kept until you delete the agent or your account.
Server logs — kept for a limited period by our infrastructure providers for security and debugging.

Your choices and rights

You can ask us to access, correct, export, or delete personal data we hold about you — including a transcript of a meeting you were in, whether or not you're a Mainroom customer. Email hello@mainroom.ai and we'll respond within 30 days. For requests about a meeting you attended but didn't organize, we may need to confirm your identity and coordinate with the account that owns the agent, since it's their record too. This includes any notes an agent has kept about you across sessions.

Depending on where you live, you may have additional rights — for example to object to or restrict processing, to withdraw consent, or to complain to a data protection authority. We honor these requests regardless of where you are.

In the product you can also: turn off an agent's memory, turn off its post-call email, change who receives that email, disconnect tools, mute an agent mid-meeting, and delete agents or whole sessions.

Security

Traffic is encrypted in transit (HTTPS/TLS), and data is encrypted at rest by our database provider.
Agent data is scoped to its owner's account at the database level.
Credentials for tools you connect stay on our servers and are never delivered to the in-meeting agent page or to other participants.
Public entry points — the demo agent and the MCP server — can only ever summon the open demo agent or a temporary body. They cannot reach agents belonging to an account.

No service is perfectly secure. If you believe you've found a vulnerability, please write to hello@mainroom.ai.

International transfers

Our providers operate in the United States and other countries, so your data may be processed outside the country where you live, including the United States.

Children

Mainroom isn't intended for anyone under 16, and we don't knowingly collect their personal data. If you believe a child's data has reached us, contact us and we'll delete it.

Changes

If we change this policy we'll update the date at the top, and for material changes we'll notify account holders by email.

Contact

Questions, requests, or complaints: hello@mainroom.ai
[LEGAL ENTITY NAME + POSTAL ADDRESS]