# Privacy Policy

This policy explains what personal data Milly Lab collects, why it is processed, who else receives it, and how you control it. It applies to the Milly Lab web application, the Milly Lab iOS and Android apps, and the Milly Lab API. It is for anyone who uses Milly Lab or is considering it.

Last updated: 2026-10-24 · <https://m-lab.app/en/docs/privacy-policy>

---

## Who we are

Milly Lab is operated by **Milly Lab LLC**, a company registered in Tajikistan.

For data-protection purposes the operator is the **controller** of the personal data described in
this policy.

Data-protection contact: the [Contacts](https://m-lab.app/en/contacts) page.

You can also write to [support@m-lab.app](mailto:support@m-lab.app).

---

## The short version

- Milly Lab is an AI workspace. What you type into it is sent to third-party AI model providers so they
  can produce a response. Those providers are listed in the Subprocessor List.
- If you connect a GitHub repository or Google Drive, Milly Lab can read files from it, and the contents
  of those files are also sent to AI model providers as context.
- Milly Lab never writes to your repository or Drive on its own. Proposed changes are held aside and are
  written only when you press Apply.
- Milly Lab does not sell personal data and does not use your content to train its own models.

---

## What data Milly Lab collects

### Data you provide when you create an account

| Data | Source | Why it is processed |
|---|---|---|
| Email address | You, or your Google account if you sign in with Google | Account identity, sign-in, transactional email, account recovery |
| Password | You (stored only as a one-way hash, never in readable form) | Authentication |
| Display name | You, or your Google account | Showing who you are in the interface and to teammates |
| Profile picture URL | Your Google account, if you sign in with Google | Display only |
| Interface language | You | Showing the interface in your language |
| Field of work ("sphere") | You, optional | Optional profile detail; not required to use Milly Lab |

### Content you submit to the service

Everything you send to Milly Lab in order to get a result:

- Chat messages and conversation history, including the model used for each message.
- Prompts for image, video, 3D and document generation, and the files those prompts produce.
- Files you upload as attachments.
- Voice recordings you make using voice input, which are converted to text.
- Swarm Mind briefs, the intermediate work produced by each model, and the final deliverables.
- Anything you write in Spaces, Teams, or a Council discussion.

### Files from a connected workspace

This is described in full in "Connected workspaces" below, because it is the most sensitive
category of data Milly Lab handles.

### Billing and plan data

| Data | Why it is processed |
|---|---|
| Your plan, subscription status and billing period | Determining what you have access to |
| Prepaid balance and transaction history | Metering usage against your balance |
| Usage records — model used, tokens in, tokens out, cost, and whether it was charged to your plan allowance or your balance, linked to the conversation | Billing accuracy, showing you your own usage, and fraud prevention |

Card payments are processed by our card payment processor, a licensed payment provider in
Tajikistan. You pay by bank card or a Tajik e-wallet on its secure payment page. The processor
receives the payment details you enter on that page and the payment amount, and processes them under
its own terms. To start a payment, Milly Lab sends it the amount, an order reference, a short
description of the purchase, and your name and email address. Milly Lab receives a payment
confirmation (the amount, the time, the bank name and a payment reference) and never sees or stores
your card number. Milly Lab records each purchase (the amount and what was bought) in your
transaction history.

### Data collected automatically

- **Authentication tokens.** After you sign in, your browser holds an access token (valid 60
  minutes) and a refresh token (valid 30 days) in browser storage: local storage, or session
  storage if you turn off "Keep me signed in". These identify you on every request.
- **Sign-in records.** Each sign-in is recorded with a coarse device label (browser and operating
  system, for example "Chrome on macOS") and a shortened IP prefix (the first two parts of an IPv4
  address). They are used to show you recent sign-ins and to email you about a sign-in from a new
  device. Up to 20 are kept per account.
- **Two-factor data, if you turn it on.** The authenticator secret, one-way hashes of your recovery
  codes, the public keys of any passkeys you add, and the browsers you choose to trust for 30 days.
- **A short-lived connection cookie.** When you connect GitHub or Google Drive, Milly Lab sets a
  single-use cookie named `connect_nonce`, scoped to `/api/connectors`, marked `HttpOnly` and
  `Secure`. Its only purpose is to confirm that the account being connected is being connected by
  the same browser that started the request. It is consumed and discarded immediately afterwards.
  Milly Lab does not use advertising or tracking cookies.
- **Where you came from, at sign-up.** On your first visit, your browser keeps the host name of
  the site that referred you (for example `chatgpt.com`) and any `utm_source`, `utm_medium` and
  `utm_campaign` values from the link, in local storage for 90 days. No full address, page path or
  IP address is kept. These values are sent once, when you create your account, and stored with it
  so Milly Lab can count which channels bring new accounts. They are deleted with your account.
- **Operational logs.** Server logs recording requests, errors and timing.
- **IP address.** Received by the hosting provider as an unavoidable part of serving a request.

### Data Milly Lab does not collect

Milly Lab does not ask for and does not want government identifiers, health data, or financial account
numbers. Do not put them into prompts. Content in prompts is transmitted to third-party model
providers, and Milly Lab cannot retrieve it from them once sent.

---

## Connected workspaces — read this section carefully

Milly Lab can connect to a GitHub account or Google Drive so that Swarm Mind runs can read your files
and propose changes to them. This gives Milly Lab access to material that is usually private, so the
following is stated precisely.

### What Milly Lab can read

**GitHub.** Milly Lab requests the OAuth scope `repo`.

> This is important: `repo` is GitHub's scope for full read and write access to repository content.
> It is **not** limited to the single repository you pick inside Milly Lab. Once you grant it, the access
> token GitHub issues is technically capable of reading and writing **every repository your GitHub
> account can access, including private ones and those belonging to organisations you are a member
> of**.
>
> Milly Lab restricts itself to the repository you select for a session, and rejects file paths that try
> to escape it. That is a restriction imposed by Milly Lab's own code, not a limit enforced by GitHub.

**Google Drive.** Milly Lab requests the OAuth scope
`https://www.googleapis.com/auth/drive.file`. This scope is limited by Google: it grants access
only to files that you specifically open with Milly Lab and files that Milly Lab itself creates. Milly Lab
**cannot** see the rest of your Drive.

### What Milly Lab does with those files

1. During a Swarm Mind run, models can list directories and read file contents from the connected
   workspace.
2. **The contents of files that are read are sent to third-party AI model providers as context**,
   in the same way your typed prompts are. If your repository contains secrets, credentials,
   personal data about other people, or material you are contractually forbidden to disclose, do
   not connect it. Review what is in a repository before connecting it.
3. Read volume is capped — roughly 48 KB per file read and 300 entries per directory listing — so
   large files are truncated rather than transmitted in full. This is a performance and cost
   measure, not a privacy guarantee.

### What Milly Lab does when it wants to change a file

**Nothing is written to your repository or Drive during a run.** Every proposed create, edit or
delete is held in a staging area attached to that session. You see the complete list of proposed
changes in a review panel.

A change reaches your real repository or Drive **only** when you press **Apply**. At that point you
choose where it goes — the default branch, a new branch, or a newly created repository. If you
discard the changes, or simply never apply them, nothing is written.

When applying to GitHub, Milly Lab checks that the file has not changed since it was read. If a
teammate has committed to that file in the meantime, that file is reported as a conflict and left
alone rather than overwritten.

### Tokens, and what happens when you disconnect

The access and refresh tokens issued by GitHub and Google are stored in the Milly Lab database. They are
encrypted at rest with Fernet symmetric encryption, using a key held in the server environment and
not in the database. In production the application refuses to start unless this encryption key is
configured, so encryption at rest does not depend on deployment configuration.

Disconnecting a workspace in Milly Lab removes the stored connection and its token.

Disconnecting in Milly Lab does not by itself revoke the authorisation on the provider's side. To revoke
it completely:

- **GitHub** — Settings → Applications → Authorized OAuth Apps → Milly Lab → Revoke.
- **Google** — [myaccount.google.com/permissions](https://myaccount.google.com/permissions) → Milly Lab
  → Remove access.

You should do this as well as disconnecting inside Milly Lab if you want the grant fully withdrawn.

---

## Why Milly Lab processes your data, and on what legal basis

| Purpose | Legal basis |
|---|---|
| Creating and running your account | Performance of a contract |
| Sending your prompts and files to AI providers to produce the result you asked for | Performance of a contract |
| Metering usage and billing | Performance of a contract |
| Transactional email — verification, password reset, team invitations | Performance of a contract |
| Keeping the service secure, preventing abuse and fraud | Legitimate interests |
| Aggregate product analytics | Legitimate interests |
| Counting which channels bring new accounts (where you came from, at sign-up) | Legitimate interests |
| Meeting legal and accounting obligations | Legal obligation |
| Marketing email, if any is sent | Consent |

Milly Lab does not carry out automated decision-making that produces legal or similarly significant
effects about you.

---

## Who else receives your data

Milly Lab relies on third parties to operate. Each one receives only what it needs. The authoritative,
maintained list — what each one receives, when, and where data is processed — is the
**Subprocessor List**, which forms part of this policy.

In summary:

| Provider | What it receives |
|---|---|
| Anthropic, OpenAI, Google, DeepSeek, xAI | Prompts, conversation history, attachments and connected-workspace file contents, for the model you chose |
| fal.ai, Runway, Meshy | Image, video and 3D generation prompts and any reference media |
| OpenAI (speech-to-text) | Voice recordings made with voice input |
| Tavily | Search queries, when web search is used for a message, and the web addresses the assistant crawls, maps or could not open directly |
| E2B | Code executed in the sandbox, when a run uses it |
| Cloudflare R2 | Generated and uploaded files, and nightly encrypted database backups |
| Sentry | Crash reports and request timings, once crash reporting is turned on: scrubbed error messages and stack traces, the request path or page address without query, and browser and device type — never request bodies, cookies or account details |
| Resend | Your email address and the contents of transactional email |
| Our card payment processor, a licensed payment provider in Tajikistan | Payment details you enter on its payment page, the payment amount, an order reference, and your name and email address |
| Railway | Hosting — all application data passes through it |
| GitHub, Google | Only where you have connected a workspace |

Different model providers apply different data-retention and training policies to what they
receive.

Milly Lab does not sell personal data and does not share it with advertisers.

Milly Lab may disclose data where legally required, or to establish or defend legal claims.

---

## Where data is processed

Milly Lab's application and database are hosted by Railway, and files are stored with Cloudflare R2.
The providers above process what they receive in their own data centres, as described in the
Subprocessor List; for most of them this includes the United States. DeepSeek processes data in
the People's Republic of China.

If you are in the UK, EU, or another jurisdiction restricting international transfers, your data
will be transferred outside that jurisdiction as an inherent part of using Milly Lab.

---

## How long data is kept

Milly Lab does not yet apply automatic retention limits. The table below describes what actually
happens today, rather than a policy that is not enforced.

| Category | Current behaviour |
|---|---|
| Account record | Kept until the account is deleted; then replaced by an anonymous placeholder |
| Conversations, messages, swarm sessions, generated assets | Kept until you delete them or your account — there is no automatic deletion. Deleting a conversation in the interface removes it. |
| Usage and billing records | Kept for accounting, also after the account is deleted, without name or contact details |
| Session event logs | Kept until the account is deleted |
| Connected-workspace tokens | Kept until you disconnect the workspace or delete your account |
| Server logs | Determined by the hosting provider |
| Data held by AI providers after Milly Lab sends it | Governed by each provider, not by Milly Lab. See the Subprocessor List. |

---

## Your rights

Depending on where you live you may have the right to access your data, correct it, delete it,
receive a portable copy, object to or restrict processing, and complain to a regulator.

### How to exercise them today

Use the [Contacts](https://m-lab.app/en/contacts) page. Requests are answered without undue delay, and within any
period the applicable law sets.

You can also write to [support@m-lab.app](mailto:support@m-lab.app).

### What the product supports

- **Deleting individual conversations** — supported in the interface.
- **Deleting your account** — Settings → Data controls → Delete account erases the account at
  once, after you confirm your identity again. Your conversations and messages, memory, Spaces,
  agents, files and generated media, schedules, workflows, connected accounts and API keys are
  deleted. Connected-account tokens are revoked with the provider where the provider allows it.
  Every session ends. Payment and usage records (transactions, usage, subscription and balance) are
  kept for accounting. Afterwards they point to an anonymous placeholder account with no name,
  e-mail address or contact details, and the billing profile keeps only the country. Stored files
  are removed in the background shortly after.
- **Exporting your data** — Settings → Data controls → Download my data gives you, straight away,
  one zip of JSON files with everything your account holds: your profile and settings, every
  conversation with its messages, and every other record that belongs to you. Secrets and
  credentials are left out. Uploaded files and generated media are listed with their links, not
  copied in.

---

## Security

What Milly Lab actually does:

- Traffic to Milly Lab is served over HTTPS.
- Passwords are stored as one-way hashes, never in readable form.
- Sessions can be revoked. Signing out and changing your password invalidate tokens issued earlier,
  so a stolen token stops working.
- Connected-workspace tokens are encrypted at rest.
- Each user's data is scoped to their own account, and requests are checked against the
  authenticated user.
- Changes proposed to a connected workspace are staged and require your explicit approval, so a
  model cannot alter your repository on its own.

Milly Lab does not hold a SOC 2 report, an ISO 27001 certificate, or any other third-party security
certification, and does not claim compliance with any security standard.

To report a vulnerability, use the [Contacts](https://m-lab.app/en/contacts) page.

Milly Lab will notify you and any relevant regulator of a personal-data breach where required.

---

## Children

Milly Lab is not intended for children. You must be old enough to use online services under the
law where you live, as the Terms of Service describe. Milly Lab does not knowingly collect data from
children below that age. If you believe a child has created an account, contact us through the
[Contacts](https://m-lab.app/en/contacts) page and it will be removed.

---

## Changes to this policy

This policy may change as Milly Lab changes. The current version is always available at
[https://m-lab.app/docs/privacy-policy](https://m-lab.app/en/docs/privacy-policy).

Material changes — in particular any change to what is sent to AI providers, or a change of
payment provider — will be notified by email to the address on your account before they take
effect.

---

## Contact

Milly Lab LLC, Tajikistan. Use the [Contacts](https://m-lab.app/en/contacts) page.

You can also write to [support@m-lab.app](mailto:support@m-lab.app).

---
