Privacy Policy
Effective Date: April 1, 2026
Last Updated: November 14, 2026
1. Introduction
This Privacy Policy describes how MF0 (mf-lab) ("Service", "we", "us", or "our") collects, uses, and protects your information when you use our platform.
By using the Service, you agree to the practices described in this Policy.
2. Information We Collect
We collect information in the following categories:
2.1 Information You Provide
- Account Information (if applicable)
Name, email address, login credentials, authentication data
- User Content
Messages, prompts, responses, uploaded files (images, documents, audio), and any data you submit through the Service
- Feedback and Communications
Messages sent to support, bug reports, or feedback submissions
2.2 Automatically Collected Technical Data
Once per day per authenticated session, we collect the following technical data. This collection is based on our legitimate interest in improving service quality and security. It does not require cookie consent.
- IP address — used to determine approximate country of origin and detect anomalies
- Time zone — browser-reported; used to verify geographic consistency
- VPN / proxy detection — best-effort heuristic comparing IP-based timezone with browser timezone; no external tracking service is used
- Interface language — the language selected by the user in the app settings
- Operating system — detected from the browser User-Agent string (e.g. Windows, macOS, Android, iOS)
- Device type — mobile or desktop, detected from the User-Agent
- Screen resolution and color depth — used to optimise the interface
- CPU core count — reported by the browser via
navigator.hardwareConcurrency
- Available RAM — reported by the browser via
navigator.deviceMemory (rounded value, in GB)
- Network connection type — reported by the browser (e.g. 4g, wifi), where available
- Referral source — the page the user came from (
document.referrer), if any
This data is stored in our database and is not shared with advertising networks. It is used exclusively to understand how the platform is used and to improve reliability and performance.
2.3 Usage Data
- Interaction Data — features used, session activity, AI model selections
- Log Data — timestamps, request metadata
2.4 Local Storage and Project Data
MF0 is designed with local-first principles:
- Conversations, memory graph data, and project data may be stored locally on your device
- Some data may also be stored in a PostgreSQL database depending on deployment settings
- Media files (e.g., images) may be stored on disk in local storage paths
2.5 Connected Gmail Data
If you choose to connect Gmail in My services, we request permission to read and modify your mailbox and create drafts. MF0 does not send email or permanently delete email. Moving a message or a selected group of up to 50 messages to Trash is performed only after your explicit confirmation and can be reversed in Gmail.
- Connection data — your Gmail address, Google account identifier, granted scopes and connection timestamps. OAuth access and refresh tokens are encrypted and stored only on our servers; they are not exposed to your browser.
- Mailbox data you request — unread status and, when you ask the AI to read your Inbox, up to 100 latest messages with their ID, sender, subject, date and plain-text body. HTML and attachments are not read. The selected AI model processes this content to answer your request.
- Action previews — the requested action, target message ID or deduplicated list of up to 50 message IDs, server-verified senders and subjects, label name or reply-draft content, and connection identifiers. A preview expires after 10 minutes and cannot be used afterward; expired previews are periodically deleted.
Gmail data is used only to provide the features you request: reading and summarizing mail, changing read/unread status, adding labels, archiving, moving messages to Trash after your explicit confirmation, and creating reply drafts. One confirmation can apply a Trash action to the entire reviewed group; per-message success and failure results are returned so a partial failure does not cause the whole group to be repeated blindly. We do not use Gmail content for advertising, lending or sale. Disconnecting Gmail deletes locally stored OAuth credentials and requests revocation from Google; you can also revoke access in your Google Account.
Raw Gmail content is processed only for the requested feature and is retained as part of the corresponding conversation until you delete it. Aggregated or anonymized service metrics, such as token counts, model selection, timestamps and request costs, are used for billing, reliability and abuse prevention. Gmail content is not used to train, retrain, fine-tune or otherwise improve any AI/ML model beyond the requesting user's personalized model.
2.6 Connected GitHub Data
If you choose to install our GitHub App, we process only the repositories you make available to that installation. Metadata remains read-only. Contents, issues, pull requests, Actions and repository creation may have write permission solely for the confirmed actions described below.
- Connection data — your numeric GitHub user ID and login, installation ID, installation account login, permissions and connection timestamps. GitHub OAuth access/refresh tokens and installation access tokens are encrypted and stored only on our servers; your browser does not receive them. The OAuth token verifies your identity and your current access to the installation, while the installation token is used for repository reads. A separate operation-scoped write token is minted only after your explicit confirmation, used for that operation and then revoked.
- Repository data you request — repository names, descriptions and default branches, file trees, code-search matches, text files, commits, merged pull requests and releases, issue and pull request conversations and reviews, pull request diffs, and workflow-run status, jobs, steps and job logs. Job logs are fetched only through GitHub-generated HTTPS redirects without forwarding your access token to the signed download URL. The selected AI model processes this content to answer your request.
- Action previews — the requested repository name or owner/repository, branches and head commits, proposed text files and before/after diffs, issue or pull request numbers, comment and review text, label changes, workflow run IDs and connection identifiers. Preview payloads are encrypted, expire after 10 minutes and are periodically deleted.
- Confirmed action results — repository, branch, commit, issue, comment, review, pull request and workflow URLs and identifiers, plus action status and timestamps, so we can show what was applied and avoid duplicate execution. Confirmed test runs also store their runner status and a bounded install/test log so you can inspect the result.
GitHub data is used only to provide the repository features you request. A repository must be accessible both to the App installation and to your connected GitHub user; if your user access is revoked, MF0 no longer treats the repository as available. After explicit confirmation, MF0 may create a private repository with an initial text-file commit, create one atomic commit on a new branch and open a pull request, add one commit to an existing same-repository mf0/* pull request branch, comment on or triage an issue, publish a pull request review, rerun a completed workflow (optionally failed jobs only), or clone a pull request for an isolated install and standard npm test run. Issue, pull request head and workflow state are revalidated before writing. Test workspaces are deleted afterward; install and test commands receive no GitHub token, database credentials or production secrets. MF0 does not merge pull requests, force-push, delete repositories or branches, change repository settings, modify workflow files or write directly to a repository's default branch. Disconnecting deletes the local identity and installation connection, cancels pending previews and requests revocation of the GitHub OAuth authorization and current installation token; you can also uninstall the GitHub App or change its repository selection in GitHub.
2.7 Connected WordPress.com Data
If you choose to connect a WordPress.com site, MF0 requests the global OAuth scope to provide full site-building and administration features. MF0 binds every operation to the site you enter and does not use the connection to manage your other sites. We do not collect your WordPress.com password.
- Connection data — numeric WordPress.com user and site IDs, site URL, granted scopes and connection timestamps. The OAuth access token is encrypted and stored only on our servers; your browser does not receive it.
- Site content you request — site name, description and URL, post or page IDs, types, titles, statuses, dates, excerpts and content, revision IDs and content, top-post and traffic statistics, media IDs, titles, URLs, types and alt text, comments and moderation status, categories, tags, navigation menus, editable settings, theme templates and reusable template parts. The selected AI model processes this content to answer your request.
- Action execution records and results — requested content and Gutenberg markup, before/after diffs, publication and deletion targets, reviewed revision state, uploaded file bytes and metadata, term and comment changes, menu items, editable settings, template and template-part changes, site launch state, export IDs and time-limited download links, connection identifiers, action status and resulting content IDs and links. Pending action payloads are encrypted, expire after 10 minutes and are periodically deleted.
- Traffic reports and schedules — your selected weekly UTC hour, site ID, report timestamps and summarized views and visitors for the last 7, previous 7 and last 30 days. Reports run only while enabled and are stored so you can read them later.
When you select WordPress and ask the agent to work on your site, MF0 executes the requested tool action immediately: creating and publishing content, editing content and metadata, restoring revisions, generating images through the existing billed image-generation provider and uploading media, moderating comments, creating terms and updating menus, settings, templates and reusable template parts, launching the site and creating an export with a time-limited download link. Image-generation prompts and resulting files are processed to provide the requested media and record the corresponding usage charge. New content is published by default unless you explicitly request a draft. There is no additional confirmation window. Current revision state is revalidated before writing to avoid overwriting a newer external edit. Permanent deletion of a post, page, media item or navigation menu is available only when your current request explicitly asks for permanent deletion; MF0 cannot undo it. MF0 does not provide the model with your OAuth token or permit arbitrary endpoint calls. Disconnecting deletes the local connection and cancels pending actions. You can also revoke the application in your WordPress.com account settings.
2.8 Connected Dropbox Data
If you choose to connect Dropbox in My services, MF0 requests access to your account information, folder and file metadata, and file contents, including permission to create, replace and move files after your explicit confirmation. MF0 does not permanently delete files.
- Connection data — your Dropbox account identifier, email address, granted scopes and connection timestamps. OAuth access and refresh tokens are encrypted and stored only on our servers; they are not exposed to your browser.
- Dropbox data you request — folder entries, file names, paths, sizes, revisions and modification timestamps, search matches, and the contents of files you ask the AI to read. Plain text, source code, PDF, Word, Excel, PowerPoint, OpenDocument and RTF files are supported, with download and extracted-text limits. The selected AI model processes this content to answer your request.
- Downloads — files up to 25 MB are streamed through an authenticated MF0 endpoint when you open a download link. Your Dropbox OAuth token is never placed in that link or sent to your browser.
- Action previews and results — the requested operation, source/destination/original paths, file bytes and revision state are encrypted and expire after 10 minutes. Confirmation rechecks the current revision and destination before creating or replacing a file or moving it to MF0 Trash. Restored files return to their recorded original path. Result metadata and action status are retained temporarily to show the outcome and prevent duplicate execution.
Dropbox data is used only to provide the folder, search, reading, download and confirmed file-management features you request. File content returned by the AI becomes part of the corresponding conversation and is retained under the same rules as that conversation until you delete it. Disconnecting Dropbox cancels pending previews, deletes locally stored OAuth credentials and requests revocation from Dropbox; you can also revoke access in your Dropbox account settings.
2.9 Connected Google Calendar Data
If you choose to connect Google Calendar in My services, MF0 requests permission to read your calendar list and read, create, update and delete your events. Event changes are performed only after your explicit confirmation. Guest invitations and update notifications are disabled by default.
- Connection data — your Google account identifier, email address, granted scopes and connection timestamps. OAuth access and refresh tokens are encrypted and stored only on our servers; they are not exposed to your browser.
- Calendar data you request — calendar names, properties and access roles, event titles, descriptions, locations, start/end times, time zones, attendees, status, links, timestamps and ETags. The selected AI model processes this content to answer your request.
- Action previews and results — the requested operation, calendar and event identifiers, proposed event content, current event state and ETag are encrypted and expire after 10 minutes. Confirmation rechecks the current ETag before updating or deleting an event. Action results are retained temporarily to show the outcome and prevent duplicate execution.
Calendar data is used only to provide the event reading, creation, update and deletion features you request. Recurring-series management is not supported in this version. Calendar content returned by the AI becomes part of the corresponding conversation and is retained under the same rules as that conversation until you delete it. Disconnecting Google Calendar cancels pending previews, deletes locally stored OAuth credentials and requests revocation from Google; you can also revoke access in your Google Account.
3. How We Use Information
We use information to:
- Provide and operate the Service
- Process AI requests and return outputs
- Maintain conversation history and memory systems
- Enable analytics (usage, token estimates, cost calculations)
- Improve functionality and performance
- Detect and prevent abuse, fraud, or security issues
- Respond to user inquiries and support requests
4. AI Processing and Third-Party Providers
The Service may send your Input to third-party AI providers (e.g., language model APIs) to generate Output.
- Requests are routed via a server-side proxy
- API keys are not exposed to the client
- Third-party providers may process your data according to their own policies
Gmail and Google Calendar content is sent only to the AI provider you select from this Limited Use-compatible list: OpenAI, Anthropic, Google paid Gemini API, Perplexity, xAI, DeepInfra, Cerebras, Z.AI and BytePlus. Their current public API terms state that they do not use your content to train models or do not retain it, with limited processing for safety, billing, legal compliance and abuse prevention. We do not send Gmail or Google Calendar content to Moonshot/Kimi while their API training terms remain unconfirmed.
The use of raw or derived user data received from Workspace APIs will adhere to the Google User Data Policy, including the Limited Use requirements. Gmail and Google Calendar content is not sold, used for targeted advertising or lending, or transferred to third parties for any purpose other than to provide or improve the features you request. Aggregated or anonymized data is not used to train AI/ML models either.
Third-party providers may process your data according to their own policies. We review these policies and restrict Gmail and Google Calendar access to compatible providers, but we are not responsible for independent changes to third-party services or their policies.
5. Analytics and Usage Tracking
We collect and store usage-related data such as:
- Model usage (tokens, requests, providers)
- Estimated cost calculations
- Feature usage metrics
This data is used for:
- Displaying analytics dashboards
- Improving system performance
- Internal monitoring
6. Data Sharing
We do not sell your personal data.
We may share data only in the following cases:
- With service providers (hosting, infrastructure, authentication)
- With AI providers to process requests
- To comply with legal obligations
- To protect rights, safety, and security
7. Data Retention
We retain data:
- As long as necessary to provide the Service
- Until you delete it (where deletion is supported)
- As required for legal or security purposes
Expired Gmail, Dropbox and Google Calendar action previews are periodically deleted. Gmail, GitHub, WordPress, Dropbox and Google Calendar content that becomes part of an AI conversation is retained under the same rules as that conversation until you delete it. OAuth credentials are deleted when you disconnect the corresponding service or delete your account.
You may:
- Delete conversations
- Clear project data and caches
- Remove stored files and attachments
8. Your Rights and Controls
Depending on your location, you may have the right to:
- Access your data
- Correct your data
- Delete your data
- Export your data
- Restrict processing
You can manage your data via:
- Application settings
- Local storage controls
- Account deletion (if applicable)
9. Security
We implement reasonable technical and organizational measures to protect your data.
However:
- No system is completely secure
- You are responsible for safeguarding your own access and environment
10. Children
The Service is not intended for children under 13.
We do not knowingly collect data from children under this age.
11. International Data Transfers
Your data may be processed in different jurisdictions depending on:
- Server location
- Third-party providers
We apply appropriate safeguards where required.
12. Changes to This Policy
We may update this Privacy Policy from time to time.
We will update the "Last Updated" date and provide notice if changes are material.
13. Contact
For questions or requests related to this Privacy Policy:
info@mf0.ai