Privacy Policy

Last updated: October 13, 2026

The privacy of your data, and it is your data, not ours, is a big deal to us. This policy lays out what we collect and why, how your data is handled, and your rights. Three promises up front: we never sell your data, we never use it for advertising, and we never use it to train AI models.

When we say "27agents", "we", or "us", we mean chrismdp ltd, the company that builds and operates 27agents.ai and the 27agents Slack assistants (such as Em). When we say "you", we mean the people and organizations that use them.

Who is responsible for your data

chrismdp ltd is the data controller for the personal data described in this policy. chrismdp ltd is a private limited company registered in England and Wales, company number 16198002, with its registered office at 16 Broad Street, Eye, Suffolk, IP23 7AF, United Kingdom. You can contact us about anything in this policy at support@27agents.ai.

What we collect and why

Our guiding principle is to collect only what we need.

Identity and access

You sign in with Slack. We receive and store your Slack workspace ID, your Slack user ID, your name, and your email address, so we know who you are, which workspace you belong to, and can show you your own settings. We store this in your workspace's customer record on Cloudflare.

Billing information

If you pay for 27agents, payment is handled by Stripe. Your card details go directly to Stripe and never touch our servers. We keep a record of the transactions and your plan for invoicing and account history.

Conversations with our assistants

Our assistants work inside Slack. We store the messages of channels and direct messages an assistant is a member of, so it can hold a conversation and refer back to it. Each channel's history is stored separately, per channel, in Cloudflare Durable Object storage. We also store the notes an assistant deliberately saves (short facts, capped in length), and any contacts and projects recorded in your workspace. A contact record is created either because you told an assistant about someone or because you exchange email with them: see how we store Google user data, below. Notes, and only notes, are additionally converted into vector embeddings (using Cloudflare Workers AI) and stored in Cloudflare Vectorize so the assistant can recall them later. Chat messages and email are not embedded.

Optional coding fleet beta

The Paolo coding fleet is not enabled for customer repositories yet. If your workspace is invited to the beta, it processes only a named repository and brief revision requested by a member of the Slack channel connected to that repository. GitHub source and working copies run in Cloudflare Containers; job artifacts are stored in Cloudflare R2. Task context and relevant source code go to the selected AI model through OpenRouter under the data collection restriction described below. Paolo posts the brief and job evidence back to the authorized Slack channel, where they are held under your workspace's Slack retention settings. When your workspace uses Paolo, he can merge approved or reviewed changes into the base branch of the repository connected to that channel. Fleet artifacts are subject to the configured retention period, initially 30 days. Contact support to erase a fleet business and its artifacts; workspace deletion does not currently trigger fleet business erasure.

Google user data (Gmail and Calendar)

This section describes exactly how 27agents accesses, uses, stores, and shares Google user data when you connect your Google account.

• How we access it. The Google user data we access is: your Gmail messages (headers, bodies and labels), the drafts we create for you, your Google Calendar events on the calendars you choose, the list of calendars you are subscribed to, and your Google account's email address. We access it only with your explicit consent, through Google's own consent screen. We request Google permissions for the assistant you connect. An assistant with no email job never asks for your email. The permissions we request let an assistant read and label your Gmail, create email drafts for you to review and send, read and change your calendar events, and list your calendars. The assistant does not send or delete email. It changes only calendar entries that have nobody else on them: it never invites anyone, never answers an invite, and never changes an event other people are on. We store the OAuth refresh token Google gives us in your workspace's record on Cloudflare, and mint short-lived access tokens from it each time the assistant reads on your behalf. We also store the email address of the Google account you connected, so we know which account the connection belongs to. The token is held server-side only and is never sent to your browser or to any analytics service.
• How we use it. The assistant reads your email and calendar to do the job you asked it to do: sorting new email into digests, flagging genuinely urgent messages, preparing your daily brief, spotting calendar clashes, creating an email draft when you ask, and answering questions you ask it in Slack. When you ask it to, it also adds, moves or removes a calendar entry that has nobody else on it. Processing is automated.

• How we store it. We do not keep a copy of your mailbox. Email is fetched from Gmail on demand, each time it is needed.

  We store a triage ledger: for each email the assistant has looked at, other than an ignored one-time code, the sender, the subject, the category it chose, and why. It also keeps the text the AI model was given for that email, which includes the body up to a length limit, for about 48 hours, so a triage decision can be checked and corrected if you give feedback. It is then deleted automatically. Beyond that, the one exception is feedback you choose to share, described under how long we keep your data.

  We store a contact record for each address you have written to yourself. It holds what the message headers say and what we counted: the name on the mail, the address, how many messages went each way, and when the first and last of them were. No part of any message body goes into it, and no decision about the person is made or stored. An address you have never written to gets no record. You can read and delete these records on your channel vault page.

  When an assistant reads a whole email thread it may keep one short note about the person who sent it, recording something the correspondence stated: their role, their company, what they asked for, or a date they named. It records what was said and never a judgment about the person. A note is shown on that contact, and you can delete it there.

  Where the assistant quotes or summarizes an email or event into a Slack conversation with you, that conversation is stored like any other, including what was quoted. Email is never written into the vector index. A note an assistant saves about a person you correspond with is a note, so it is embedded like every other note.

• How we share it. To read and triage on your behalf, email and calendar content is sent to the AI model doing the reading. Model calls are routed through OpenRouter with provider data collection denied: every request tells OpenRouter to only use model providers that state they do not retain or train on the content. If you send a problem report, the email or calendar content in it, and any optional context you approve, is reviewed by our own operations agent, which reaches AI models through OpenRouter on these same terms. No Google user data goes to analytics: our telemetry never includes the content of your messages or email, and this is enforced in the code that sends telemetry, which removes content fields before anything leaves our systems. There is no setting that can turn content capture on. We never sell Google user data, never use it for advertising, and never transfer it to anyone else except the subprocessors listed below, as needed to provide the service.
• Human access. Nobody at 27agents reads your Google user data in the normal course of business. The only exceptions are: with your explicit permission, which includes the feedback you deliberately submit in the private 27agents inbox and any optional context you approve for that report, and helping with a support request you made; for security purposes, such as investigating abuse; or when we are required to by applicable law.

27agents' use of information received from Google APIs will adhere to the Google API Services User Data Policy, including the Limited Use requirements. You can read that policy at developers.google.com/terms/api-services-user-data-policy.

AI processing

The assistants are powered by large language models. Every model call is routed through OpenRouter, and every call denies provider data collection, which restricts routing to model providers that state they do not retain or train on your content. Embeddings are computed with Cloudflare Workers AI inside our own Cloudflare account.

Product analytics and telemetry

We use PostHog to understand how the product is used and to monitor that it is working. What we send is metadata only: event names, user and workspace IDs, timings, token counts and costs, model names, tool names, and error codes. It never includes the content of your messages, email, notes, or the assistant's replies. This is not a policy toggle: the code that sends telemetry strips content fields before anything is sent, so there is no configuration in which content reaches our analytics. On the website, PostHog records standard usage analytics such as pages visited.

Cookies and similar technologies

When you sign in to 27agents.ai we set a signed session cookie so you stay signed in. This cookie is strictly necessary for the service to work and expires on its own. It is not part of the analytics consent banner, and accepting or rejecting analytics does not change it.

Separately, our analytics (PostHog) stores a first-party identifier in a cookie and in your browser's local storage to tell visits apart. We ask first in the banner, and analytics runs only after you accept. Saying no is as easy as accepting, and the site works the same either way. Your choice lives in this browser and can be changed at any time with the toggle on this page.

When analytics is on, we use it only to produce aggregate statistics about how the site is used, such as how many visitors each page gets and how many complete each step of signing up. Browsing on its own is not linked to your name or email address, no profile of your visits is built, and nothing is used for advertising. Two actions are identified separately from browsing: when you sign in, we record which workspace signed in, and when you join a waitlist, we store the email address you gave us so we can contact you. The website analytics events for those actions run only after you accept.

Changing the toggle to off takes effect immediately: analytics stops and the analytics identifier is deleted from your browser. The only thing kept is the choice itself, stored in your browser's local storage so it holds on return visits in this browser. We run no advertising, set no advertising or third-party tracking cookies, and never share cookie data with ad networks.

Voluntary correspondence

When you email us with a question, we keep the correspondence, including your email address, so we have a history to refer to if you contact us again.

Email from us

The first time you sign in with Slack, we ask once whether you want email from us. We save your answer on your member record. If you say yes to either kind of email, we send your email address and first name to Kit, the service we use to send email. 27agents news is occasional email about the product. The AI Leaders newsletter is a weekly email from Chris Parsons, who runs 27agents. If you ask for it, Kit first sends you an email to confirm. Every email has an unsubscribe link, and you can also ask us at support@27agents.ai to stop. If you join a waitlist, we send the address you give us to Kit too, so we can tell you when it opens.

Our lawful bases for processing

UK data protection law (the UK GDPR) requires us to have a lawful basis for each way we use your data. Here is the basis for each purpose:

• Running the service you signed up for: your account and member record, conversations with the assistants, notes and their embeddings, contacts and projects, and billing and plan records. The lawful basis is performance of a contract (Article 6(1)(b)): we process this data because it is what providing 27agents to you is.
• Google user data (Gmail and Calendar): the lawful basis is your consent (Article 6(1)(a)), which you give on Google's own consent screen when you connect your account. You can withdraw that consent at any time, and withdrawing it is as easy as giving it: use the switches on your dashboard, or revoke 27agents' access from your Google Account permissions page, as described under "Revoking access and deleting your data" below. Withdrawing consent does not affect the lawfulness of processing that happened before you withdrew it.
• Product analytics and telemetry (metadata only, described above): our legitimate interests (Article 6(1)(f)) in understanding how the product is used, monitoring that it is working, and improving it. We have deliberately limited this data so that it never includes the content of your messages or email.
• Security, abuse prevention, and support: our legitimate interests (Article 6(1)(f)) in keeping the service and its users safe, and in answering you when you contact us.
• 27agents news: our legitimate interests (Article 6(1)(f)) in telling people who use 27agents about the product. Under the UK's electronic marketing rules (PECR), we send it only because you gave us your address when you signed up, we offer the chance to refuse it when we collect the address, and every email lets you unsubscribe.
• The AI Leaders newsletter, and waitlist email: your consent (Article 6(1)(a)). You can withdraw it at any time with the unsubscribe link.
• Records we are required to keep, such as tax and accounting records of payments: legal obligation (Article 6(1)(c)).

How long we keep your data

• Conversations, notes, embeddings, contacts, projects, and member records: kept for as long as your workspace uses the service, and deleted when you ask us to delete your workspace's data. There is no automatic expiry.
• The email triage ledger (sender, subject, category, reason, and the text the AI model was given for that email, which includes the body up to a length limit): entries are routinely deleted about 48 hours after the email was first seen, once it has been included in a digest. We keep that text so a triage decision can be checked and corrected if you give feedback. Rows needed to resume deferred email triage are kept until that work completes.
• Problem reports you share for review: forward an assistant message to the private 27agents inbox and tell us what was wrong. The forwarded message and your notes are submitted for review. You can also press Share to share the conversation and context used for that reply with us. When a reply fails, we file a report with the failure details and none of your conversation, and ask in the same inbox whether to share the conversation behind it. Press Always share, or choose it under Sharing with 27agents on your dashboard, and later reports and failures are shared without asking, each still only the conversation behind that one reply. You can change that choice there at any time. That optional context can include recalled notes and tool results and, for a digest or an alert, the underlying emails as they were classified, subject and body included. We keep that content until you delete the feedback item on your channel vault page. If a report leads to a test case, we rewrite it as an invented example that keeps the shape of the problem and none of your content. Where an older test case still holds a copy, deleting the item removes that copy from our evaluation set too. We use it only to check and improve how the assistant classifies and answers, never to train a model. Press Don't share, or choose Never share, and we keep the forwarded message, your report notes and existing non-content metadata, with no original conversation context.
• What the assistant sent the model to write a reply (its instructions, the conversation so far, and the notes it recalled, as one body): kept for 24 hours, so that a problem report you send shortly after a reply can be checked against what the assistant was really working from. It is deleted automatically the next time that channel is written to or looked up, and a channel that goes quiet can hold its last one a little longer than that, but nothing older than 24 hours is ever released, to us or to anyone. Nothing is read from it at all unless you approve sharing optional context for a report.
• How many emails we processed for you each day (a count per person per day, with nothing about the emails themselves): kept for 90 days and then deleted automatically the next time this data is written to or looked up. A workspace that stops processing email, with nobody checking its usage again, can keep its last counts a little longer than that. We use it to see whether a plan's daily limit fits how you actually work.
• The Google OAuth token: kept until you revoke the connection or ask us to delete your data, whichever comes first.
• Email choices and the address sent to Kit: your choice stays on your member record as long as the account exists. Kit keeps your address until you unsubscribe or ask us to delete it.
• Sign-in session cookies: expire on their own, after at most 180 days.
• Analytics and telemetry events: kept in PostHog, our analytics service, hosted in the European Union, under its cloud retention policy of at most seven years. These events contain metadata only, never content.
• Billing records: kept for as long as tax and accounting law requires.
• Support correspondence: kept while it is useful for helping you, and deleted on request.

Automated decision-making

The assistants automatically summarize conversations and suggest what needs attention. Em classifies email, flags urgent messages, and highlights calendar clashes and meetings it considers important. These results can be wrong and can affect which messages and commitments you notice first.

Em can star messages it classifies as urgent and mark processed messages as read, according to your saved settings. You can change these choices independently. The assistant does not ask you to approve each classification or enabled email change. You can review the original messages and change their stars or read status in Gmail.

Em can also hold back an email it judges to be likely spam or fraud from a sender who does not appear in your sent mail. A held-back email gets no alert and no line of its own in the digest. Em leaves a held-back email untouched in Gmail: it is not moved, starred or marked as read. The next digest says how many emails were held back, and for 48 hours after that digest you can ask Em which ones they were.

Em ignores an email that mainly gives you a one-time code, such as a sign-in code or a code to give a delivery driver: it leaves the email untouched and unread in Gmail and does nothing else with it. Em does not count it in the digest, list it when you ask, or keep its sender, subject or contents.

When an email is work for another assistant you have added, such as Miriam, Em can pass it to that assistant. For Miriam, that means email about your goals, priorities, commitments or planning your week. That assistant tells you about it in your own direct messages with it. It is told who sent the email, the subject and why it was passed on, never the body.

Em can write an email draft in Gmail for you to review and send. The assistants do not send or delete email. When you ask, an assistant can add, move or remove a calendar entry that has nobody else on it. It never invites anyone, never answers an invite, and never changes an event other people are on, and nobody is ever notified of a change it makes. You decide whether to act on their summaries and suggestions. For questions or concerns about automated processing of your information, contact support@27agents.ai.

Subprocessors

We use these third-party services to run 27agents. Each one receives only what it needs:

• Cloudflare: hosting, storage (Durable Objects and Vectorize), and embeddings. All application data lives here. Our use is covered by the Cloudflare Customer Data Processing Addendum, which forms part of Cloudflare's customer agreement.
• Slack: the chat surface the assistants work in. Messages the assistant posts are, like all Slack messages, stored by Slack under your workspace's own Slack agreement. We have entered into Slack's data processing addendum.
• Google: the provider of your connected Gmail and Calendar account. Your mailbox and calendar stay in your own Google account, under your own agreement with Google. Our access runs through Google's APIs, whose terms of service incorporate the Google Controller-Controller Data Protection Terms.
• OpenRouter: AI model routing. We deny provider data collection on every request, and our account is set to zero data retention, so OpenRouter only routes to model providers it designates as zero-data-retention. Feedback you deliberately submit, and optional context you approve, is reviewed by 27agents' own operations agent. That agent runs on AI models reached through OpenRouter on these same terms. Our use is covered by OpenRouter's data processing agreement, which forms part of its terms.
• PostHog: product analytics and telemetry, metadata only, never message content, hosted in the European Union. We have entered into PostHog's data processing agreement.
• Stripe: payments and billing. Our use is covered by the Stripe Data Processing Agreement, which forms part of Stripe's services agreement.
• Kit (kit.com): sends our email. It receives your email address, first name and which emails you chose, and only if you chose at least one. Our use is covered by Kit's data processing addendum, which forms part of its terms.

When we access or disclose your information

Nobody at 27agents looks at your content in the normal course of business. The exceptions: to help you with a support request you made, with your express permission first; when an automated process fails partway and fixing it requires looking at the minimum amount of data needed; to investigate abuse of the service; or when required by applicable law with a legally binding order. Where the data involved is Google user data, any such access also stays within Google's Limited Use requirements described above.

If 27agents were ever acquired by or merged with another company, we would notify you well before any of your information became subject to a different privacy policy.

Revoking access and deleting your data

• Turn access off in the product. Your dashboard at 27agents.ai/dashboard has switches that stop the assistant reading your email and calendar immediately, without disconnecting anything.
• Revoke the Google connection entirely. Visit your Google Account permissions page and remove 27agents. This invalidates the token we hold, and the assistant loses all access to your Google account.
• Remove the assistant from Slack. Removing the app from a channel or your workspace stops it seeing those messages.
• Delete the workspace yourself. The workspace owner can delete the workspace from the Team page in the dashboard. This erases the workspace data we hold: conversations, notes, embeddings, the email triage ledger, member records, and connected-service credentials. We retain workspace creation and deletion history and billing records. Limited authentication records expire automatically. Deleted data cannot be restored. Once deletion finishes, you can reinstall the apps in this Slack workspace and start fresh. If you would rather ask us, or the owner has left, email support@27agents.ai and we will carry the deletion out for you, promptly.

You also have the right to know what personal information we hold about you, to access it, to have it corrected or erased, to restrict how it is processed, to receive a copy in a portable form, and, where processing is based on your consent, to withdraw that consent at any time. Email us and we will help.

Your right to object. Where we process your data based on our legitimate interests (such as the metadata-only analytics described above), you have the right to object at any time. Email us and we will stop unless we have compelling legitimate grounds to continue.

If you are unhappy with how we have handled your data, you have the right to complain to the UK supervisory authority: the Information Commissioner's Office (ICO), Wycliffe House, Water Lane, Wilmslow, Cheshire, SK9 5AF, United Kingdom, ico.org.uk. chrismdp ltd is registered with the ICO under registration number ZB967439. We would appreciate the chance to help first, but you can go to the ICO at any time.

How we secure your data

All data is encrypted in transit with TLS. Data at rest on Cloudflare's network is encrypted by the platform. OAuth tokens are held server-side only. Access to production systems is restricted to the people who operate the service.

Where your data is processed: international transfers

27agents runs on Cloudflare's global network, and most of our subprocessors are US companies, so your data may be stored and processed in the United States and other countries outside the UK. One exception by design: our analytics project is hosted by PostHog in the European Union (Frankfurt, Germany), so analytics data does not leave the UK and EU. Where personal data does leave the UK, we rely on the safeguards UK data protection law recognizes:

• Cloudflare, Google, Slack (through Salesforce), Stripe, PostHog, and Kit are each certified under the EU-US Data Privacy Framework, including its UK Extension (the UK-US "data bridge"), which UK law recognizes as adequate protection for transfers to those companies in the United States.
• OpenRouter is not certified under the Data Privacy Framework. Model calls through OpenRouter are transient by construction: OpenRouter does not use your content for training, and we deny provider data collection on every request, which restricts routing to model providers that state they do not retain your content. Transfers to OpenRouter are governed by its data processing agreement, which incorporates the standard contractual clauses approved by the European Commission (module two, controller to processor) and applies the UK Addendum to transfers originating in the United Kingdom.

You can ask us for a copy of the safeguards we rely on, and a data processing agreement is available on request.

Changes and questions

We may update this policy as the product and regulations evolve. When we make a significant change, we will refresh the date at the top of this page and take appropriate steps to notify you. Questions about this policy or your data? Email support@27agents.ai.