Trust center

How we look after your data, and where our compliance work stands.

How your data is protected

Encrypted in transit and at rest. Mailbox and connector credentials are encrypted again inside the database, API keys are kept only as hashes, and we don’t train models on customer data. Everyone can see where they’re signed in and end any session, and an admin can sign out browsers left idle and make every sign-in expire after a set time.

Every page is served over HTTPS alone, and a browser that has been here is told to ask for nothing else for two years. No other site can show the app inside a frame of its own.

Signing in

  • Two-step verification, which each person turns on under Settings, Security: a six-digit code from an authenticator app after the password, or one of ten backup codes, each good once and stored encrypted. An emailed sign-in link asks for the code too; signing in with Google is left to Google’s own second step.
  • Single sign-on on Enterprise, through your own identity provider over SAML 2.0 or OpenID Connect, for an email domain proved with a DNS record. Required, it turns away passwords, emailed links and Google for addresses at the domain, except an owner’s password, kept so a broken identity provider can’t lock the workspace out.
  • Guessing is stopped: ten wrong passwords for one address within fifteen minutes stop it taking passwords for up to fifteen minutes, though an emailed sign-in link still works, and ten wrong codes in a row lock the second step for fifteen minutes.
  • Sign-ins end: in the browser, one left idle ends after 2 hours and any ends after 30 days, unless an admin sets other limits, from 15 minutes to a day idle and from 8 hours to 30 days in all. The phone apps stay signed in until somebody signs them out, in the app or on the Security screen.
  • Roles: everyone in a workspace is an owner, an admin, a member, or Pages only, who writes and reads Pages without seeing an account, a deal or a call. A page can be shared with view, edit or full access.

AI and your data

Migaru’s AI features send the data a task needs, such as a record, a transcript or a message snippet, to a large language model to produce the result. Models are provided by Anthropic and reached through Vercel’s AI Gateway, which sends a request only to a provider that has agreed with Vercel to keep none of it and not to train on it, and refuses the request when none has. When the assistant runs an analysis, it does so in an isolated Vercel sandbox with no network access, over data you can already see.

On plans that include web research, the assistant can search the web through Exa, reached through Vercel’s AI Gateway. It sends a search question, such as a company’s name and website or a person’s name and company, and is told to put nothing from your mail, meetings or notes into it.

No email goes to a customer until a person has approved it. Follow-ups and sequence steps are drafted for review, an approved email goes to the sender’s own mailbox for them to send, and a cold campaign sends only once somebody has written and launched it.

Every change to a record’s fields, the AI’s included, is kept in that record’s history, with what the field said before and what changed it: a person, the agent, a sync, an import or a workflow.

Who processes data for us

Only the providers that run Migaru for us, each for that purpose alone: Vercel, Neon, Anthropic, Stripe, Resend, Google, Recall.ai, Twilio, Exa and Lusha. The privacy policy says what each one does and receives, and what goes to the tools a workspace connects, and the data processing agreement where each one is. Workspace owners are told by email at least 30 days before a new one starts.

Your data stays yours

An owner or admin can export the whole workspace from Settings → General: every record, the links between records, field history, files, call transcripts and recordings, and pages, as CSV, JSON and Markdown in zip files, with every recording a download of its own. It leaves out only what a member keeps from the rest of the workspace, their private pages, pages shared only with the people they invite and the calls they share as who and when alone, and lists what it left out. The downloads last 7 days, and can be asked for again. Any view also exports on its own as CSV, Excel or JSON, and every record can be read and written through the API and the MCP server, which Claude, ChatGPT and other assistants sign in to.

  • Records stay in your workspace until you delete them. A deleted record moves to Settings → Deleted, where it can be restored or deleted for good.
  • When a workspace’s subscription ends, its mailboxes and calendars stop syncing, its campaigns stop sending and the notetaker stops joining its calls, and it stays read-only for 90 days, so its people can still read and export everything. Paid for again in that time, it picks up where it stopped. Within 30 days after those 90 days, it is deleted as an owner’s deletion does it, and its owners are emailed when it is.
  • Disconnecting a mailbox, calendar or Slack stops Migaru from reading it. It does not remove the records already filed from it; delete those in the workspace if you want them gone.
  • You can delete your account under Settings, Profile. It goes at once, with your sign-ins, connected mailboxes, keys and private pages; the records you kept for your workspace stay with the workspace.
  • An owner can delete a whole workspace under Settings, General. It is locked and its billing stops at once, and 14 days later it is deleted for good, with its records, files and the accounts of the people in it; any owner can keep it until then. Export what you want to keep first: while it waits, it opens for nobody but an owner keeping it. If you would rather we did it, email support@migaru.ai from the address on the account.
  • Deleted data is gone from our database backups within [Neon history retention, e.g. 7 days] of being deleted.

If Migaru shuts down

  • You hear first: we email every workspace owner and admin, and say so in the app, at least 90 days before Migaru stops. Until then the workspace keeps working as it always has.
  • You can take everything: after the last day, every workspace stays read-only for 90 days. An owner or admin can still export the whole workspace, any view still exports as CSV, Excel or JSON, and every record can still be read through the API and the MCP server.
  • Billing stops: nothing renews once notice is given, and time paid for past the last day is refunded.
  • Then it is deleted: within 30 days of those 90 days ending, every workspace is deleted as an owner’s deletion does it, with its records, files, call recordings and transcripts, the credentials of its connected mailboxes, calendars, Slack and Zoom, and the accounts of the people in it, and its owners are emailed when it is. Recall.ai is asked to delete what it holds of recorded calls, and our database and file storage are deleted with their backups. Only invoices and payment records outlast it, for as long as the law requires them kept.
  • It is not sold off: the data in a workspace is not sold as an asset of Migaru. If another company takes the service on instead, we tell owners before anything moves, that company is bound by this policy, and a workspace that would rather not go can be exported and deleted first. Data from Google accounts moves only with each person’s explicit consent.

The terms of service are where these, and what happens when a subscription ends, are promised.

Reliability

On Enterprise, the service is available at least 99.9% of every month, with a credit against that month’s fees when it is not: 10% below 99.9%, 25% below 99.0%, 50% below 95.0%. The terms say how it is measured.

The database can be restored to any point in the last [Neon history retention, e.g. 7 days].

If something isn’t working, email support@migaru.ai and we’ll tell you what we know, as we know it.

Compliance

  • SOC 2 Type II — audit in progress
  • HIPAA — in progress
  • GDPR
  • ISO 27001 — planned

Neither the SOC 2 Type II report nor the HIPAA work is finished yet — when they are, this page will say so and carry them. In the meantime, ask us where each one stands and you will get the real answer rather than a badge.

Migaru’s use and transfer of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements.

Incidents

If a breach of security affects a workspace’s data, its owners hear from us without undue delay, and within 72 hours of our becoming aware of it: what happened, what it may affect, and what we are doing about it, with more as we learn it.

Reporting a vulnerability

If you think you have found a vulnerability in Migaru, write to security@migaru.ai, with what you found and how to reproduce it. We reply within two business days, keep you told as we fix it, and credit you when it is fixed, if you would like.

We will not pursue or report anyone who looks for vulnerabilities in good faith and:

  • tests only against a workspace of their own, and reads, changes or keeps no one else’s data beyond what shows the problem;
  • does not degrade the service for anyone, by load, spam or otherwise, and does not try to trick our people or our customers;
  • gives us a reasonable time to fix it before telling anyone else.

Questions

Questions about security or compliance: security@migaru.ai. About anything else: support@migaru.ai.

Updated