Privacy Policy
Magister Marketing AI Agent Platform
Effective date: March 14, 2026 · Last updated: November 12, 2026
Overview
Magister Marketing ("Magister," "we," "us") is an autonomous AI marketing platform. Every brand on every plan, including the free trial, has its own isolated hosted workspace on a dedicated cloud machine; the Agent plan and the free trial also run Magister's own agent on it. Depending on the plan and features you enable, Magister can manage marketing tasks, interact with third-party services, send communications, and search the web. This privacy policy explains what data we collect, how we use it, who we share it with, and what rights you have over your data.
Data We Collect
Account Information
When you create an account, we collect your email address, authentication credentials managed by Supabase (we never store or access plaintext passwords), display name, and avatar URL.
Billing Information
Payments are processed by Stripe. We store your Stripe subscription ID, plan tier, and subscription status. We do not store credit card numbers, bank account details, or other payment credentials — Stripe handles all payment data directly.
Chat & Conversations
We store the full conversation history between you and your AI agent, including message content, timestamps, and which AI model generated each response. This data is stored in Supabase and is scoped to your account.
Workspace Machine Data
For each brand, on the Connect plan, the Agent plan and the free trial, we provision an isolated machine on Fly.io that holds the brand's hosted workspace (and, on the Agent plan and the free trial, runs Magister's own agent). We store the Fly.io app and machine IDs, region, status, and activity timestamps associated with that machine. While a machine is paused for being idle, Fly.io keeps its disk, and may keep a snapshot of its memory, so it can resume.
Usage Data
We track LLM token counts, costs, and the model used for each request. This data is used for billing enforcement and plan limits.
Integration Tokens
When you connect third-party services (including Google, GitHub, Vercel, Webflow, Wix, WordPress, Shopify, PostHog, Slack, and supported social or advertising accounts), we store the resulting access and refresh tokens. DataForSEO credentials are stored if you provide them. All integration tokens are encrypted at rest using Fernet symmetric encryption.
External AI Clients and MCP Connectors
When you authorize an external AI client such as Claude, ChatGPT, Codex, Cursor, or Grok through Magister's MCP connector, that client can request the Magister and connected-service data covered by the organization and scopes you approve. With the default read scope this includes your brands' past Magister chat history (the same conversations members of that brand see in the app), workflow run history, permission requests, and the record of what the agent drafted or shipped. Tool responses are returned to that client and are also subject to the client provider's privacy terms. Separately scoped tools can send bounded request data to the server-side research and integration providers listed below or initiate writes you request. Magister continues to enforce its server-side scopes, roles, budgets, and each tool's documented immediate-confirmation requirement. Connected-service credentials remain encrypted on Magister's servers and are never returned to the external AI client.
We never receive or reconstruct your conversation with that client. We record each tool call Magister handles: the tool, its outcome, the client, a summary of the request, and, when the client provides them, a short one-sentence statement of the task you asked it to do and a conversation identifier supplied by the client. We use these to show your team what the assistant did in Magister and to improve the service. Usage events (tool name, timing, errors, and client, never tool inputs or results) are also sent to our product analytics provider, PostHog.
BYOK API Keys
On a paid Agent plan, if you bring your own API keys for LLM providers (Anthropic or OpenAI), we store them encrypted at rest. These keys are used to route your AI requests directly to the provider of your choice, billed to your own account, instead of through our platform.
Email Data
When you enable agent email features, emails sent and received through the agent can include sender, recipient, subject, and body. This data is stored to enable the email workflows you request.
Analytics
Fathom Analytics counts page views on our public marketing pages. It sets no cookies and keeps no identifier, so it runs for every visitor.
TracerKit, Headstream and Google Analytics measure how visitors find and use our site, and set cookies to recognize a returning browser. On our public marketing pages, TracerKit records the page you arrive on, the site that sent you, any ad click identifier in the link (such as gclid or fbclid), and the Meta and Google Analytics browser identifiers when those are set. It also loads Headstream, which records the pages you view, where you click and how far you scroll. Neither TracerKit nor Headstream receives what you type into a form. Google Analytics records page views across the site and the app, with private details removed from page addresses. If you are in the European Economic Area, the United Kingdom or Switzerland, these three load only after you agree. Elsewhere they are on by default and you can opt out at any time; we also treat a Global Privacy Control signal from your browser as an opt-out. You can change your choice at any time under Cookies below. When you opt out, we delete their identifiers from your browser.
We use PostHog for product analytics across the site and the app: page views and product interactions, linked to your account once you sign in, so we can understand how Magister is used and improve it. PostHog also records sessions so we can see where people get stuck; recordings mask everything typed into a form field and your chat messages, and leave out images and video in chat. PostHog runs for every visitor and is not affected by your analytics cookie choice.
Cookies
You can turn analytics and advertising cookies on or off at any time.
Some cookies make the site work: Supabase session cookies keep you signed in, and others remember the organization and brand you are working in, your sidebar layout, an offer, invitation or referral link you followed, and short-lived security values while you connect an integration. We also set mm_attr, which remembers how you first reached our site (campaign parameters such as utm_source, the referring site, and the page you landed on) so we can attribute signups to their source; it holds no identifier and lasts up to 180 days. mm_consent_region records only whether your region requires consent (30 days), and mm_ad_consent and mm_analytics_consent remember your advertising and analytics choices for a year if you make them.
PostHog stores an identifier in a first-party cookie and in local storage. Before it loads, we set mm_ph_did, which holds the same kind of identifier for up to a year, so an experiment shows you the same version on every page. When analytics is allowed, TracerKit sets tk_aid, an identifier that lasts a year, and tk_sid, which lasts 30 minutes, and keeps the identifier and any ad click identifier in local storage. Headstream sets hs_vid, an identifier that lasts about 400 days, keeps it in local storage too, and keeps a session identifier in session storage. Google Analytics sets _ga cookies that last up to two years. The TracerKit, Headstream and Google Analytics cookies are set for magistermarketing.com, so its subdomains receive them too. The Meta Pixel's cookies are described below.
Advertising Cookies
We use the Meta Pixel to measure how well our ads work: it records page views, signups, checkout starts and purchases, and Meta sets its own cookies to do so. If you are in the European Economic Area, the United Kingdom or Switzerland, the pixel loads only after you agree. Elsewhere it is on by default and you can opt out at any time; we also treat a Global Privacy Control signal from your browser as an opt-out. You can change your choice at any time under Cookies above.
When advertising tracking is allowed, we also measure signups with a server-side Meta Conversions API event. This can include a hashed email address and account ID, your IP address, browser user-agent, and Meta browser and click identifiers when available. Meta uses these values to match and deduplicate the signup event for advertising measurement. Your advertising cookie choice applies to these events too.
How We Use Your Data
- To provide and operate the Magister platform and your personal AI agent
- To process your chat messages through AI language models
- To execute marketing tasks on your behalf via connected integrations
- To process payments and enforce plan limits
- To send transactional emails (account confirmation, billing receipts)
- To monitor usage for billing and abuse prevention
- To understand how our site and the service are used, and improve them, as described in Analytics above
- To measure our advertising as described in Advertising Cookies above
We do not sell your data. Our advertising measurement uses the data described above and follows your advertising choice. We do not train AI models on your conversations.
Third-Party Services
We share data with the following third-party services as necessary to operate the platform:
| Service | Data Shared | Purpose |
|---|---|---|
| Supabase | All user data (encrypted at rest) | Database and authentication |
| Fly.io | Gateway requests, agent conversations, workspace files, and machine provisioning commands | AI agent infrastructure |
| Stripe | Billing events | Payment processing |
| Anthropic | Chat prompts and conversation context | Claude model inference (no training on API data per Anthropic Commercial Terms §B) |
| OpenAI | Chat prompts and conversation context | GPT model inference (no training on API data per OpenAI API Data Usage Policy) |
| OpenRouter, Inc. | Chat prompts and conversation context | LLM request routing through zero-data-retention endpoints only |
| Google Cloud Vertex AI | Chat prompts and conversation context | Gemini inference through zero-data-retention endpoints via OpenRouter |
| Fireworks AI | Chat prompts and conversation context | GLM inference through zero-data-retention endpoints via OpenRouter |
| Together AI | Chat prompts and conversation context | GLM inference through zero-data-retention endpoints via OpenRouter |
| Parasail | Chat prompts and conversation context | GLM inference through zero-data-retention endpoints via OpenRouter |
| Novita AI | Chat prompts and conversation context | GLM inference through zero-data-retention endpoints via OpenRouter |
| External AI clients you authorize | MCP tool requests and the authorized Magister or connected-service data returned by those tools | Using Magister from Claude, ChatGPT, Codex, Cursor, Grok, or another MCP client you choose |
| Google (Gemini media) | Image/video prompts and generated media | Gemini image and video generation |
| Sentry | Error reports, which can include request details and message content, with credentials removed | Error monitoring |
| Resend | Email content and recipients | Transactional email delivery |
| Brave Search | Search queries | Web search (agent capability) |
| DataForSEO | Keyword queries and public website domains | Keyword, ranking, search-volume, and SEO research |
| ScrapeCreators | Selected social channel, topic, result limit, and public creator identifiers | Influencer discovery and public contact evidence |
| Bright Data | Public website URLs, public-page requests, and configured AI-visibility queries | Website retrieval and public search or AI-visibility measurement |
| Zernio | Connected social and advertising account identifiers, content, campaign settings, and analytics requests | Social publishing, account analytics, and paid-ad management when you connect those accounts |
| Composio | Connected Gmail authorization and email delivery requests | Sending email through the Gmail account you connect |
| Fathom Analytics | Page views (anonymous, no cookies) | Website analytics |
| TracerKit | On our public marketing pages: the page you arrive on, the referring site, ad click identifiers in the link, a browser identifier, and the Meta and Google Analytics browser identifiers when set (with your consent where required) | Measuring which sites and campaigns bring visitors |
| Headstream | On our public marketing pages: page views, clicks, scroll depth and a browser identifier (with your consent where required) | Website analytics |
| Google Analytics | Page views across the site and the app, with private details removed from page addresses, and a browser identifier (with your consent where required) | Website analytics |
| PostHog | Page views, product interactions and session recordings across the site and the app, linked to your account once you sign in | Product analytics |
| Google (Ads, Analytics, Search Console, Tag Manager, Drive, Calendar) | OAuth tokens and API requests on your behalf (see “Google User Data” section below) | Google integrations (only when you connect them) |
| Meta | Page views, signups, checkout starts and purchases from our website (with your consent where required) | Measuring our advertising |
| GitHub, Vercel, Webflow, Wix, WordPress, Shopify, PostHog, Slack | API calls on your behalf | Third-party integrations (only when you connect them) |
Third-party integrations are activated only when you explicitly connect them or invoke a separately scoped built-in provider feature. Your agent accesses connected services with the permissions you grant, while metered research providers receive only the bounded request data needed for the tool you invoke.
Google User Data
This section describes how Magister handles data received from Google APIs in compliance with the Google API Services User Data Policy, including the Limited Use requirements.
Google scopes we request
When you connect a Google account in Settings → Connections, Magister requests only the minimum scopes required to provide the features you asked for:
- Google Ads (
adwords) — lets your agent manage ad campaigns, read performance reports, and adjust bids, keywords, and ad copy on your behalf. - Google Analytics 4 (
analytics.readonly) — lets your agent query read-only GA4 performance data to help you track the impact of marketing changes. The agent never modifies GA4 configuration. - Google Search Console (
webmasters.readonly) — lets your agent read search performance, top queries, and coverage data for your verified sites. - Google Tag Manager (
tagmanager.readonly) — lets your agent audit tag configuration and verify that analytics and conversion tracking tags are correctly published. Read-only; never modifies GTM resources. - Google Drive (
drive.file) — lets your agent create new Docs, Sheets, Slides, and folders on your behalf. This is a per-file scope: the agent can only access files it creates, never your existing Drive contents. - Google Calendar (
calendar) — lets your agent schedule marketing events, content deadlines, and reminders when you explicitly request it.
How we store Google user data
Google OAuth access and refresh tokens are stored encrypted at rest in our Supabase database using Fernet symmetric encryption. Tokens are never written to your agent machine or any other runtime environment. Our gateway injects the real token server-side at request time and strips it from the response path, so the agent runtime never sees your Google credentials.
How long we retain Google user data
We retain Google OAuth tokens and any metadata fetched from Google APIs only for as long as the integration is connected. When you disconnect a Google integration from Settings → Connections, the associated tokens and metadata are deleted immediately. When you delete your Magister account, all Google user data is permanently removed.
Limited Use of Google user data
Magister's use and transfer of information received from Google APIs to any other app will adhere to the Google API Services User Data Policy, including the Limited Use requirements. Specifically:
- We do not sell Google user data to any third party.
- We do not use Google user data for advertising, including retargeting, personalized advertising, or interest-based advertising.
- We do not use Google user data to train, fine-tune, or improve generalized or non-personalized AI/ML models.
- We do not allow humans to read Google user data except with your explicit consent, for security investigations, to comply with applicable law, or where the data has been aggregated and anonymized for internal operations.
- We do not share Google user data with any third party except to provide the features you have explicitly asked your agent to perform (for example, forwarding your request to Google's own APIs).
How to revoke access to Google user data
You can revoke Magister's access to your Google account at any time from two places:
- Inside Magister: Settings → Connections → Disconnect. This immediately deletes the OAuth tokens and associated metadata from our database.
- From your Google account: https://myaccount.google.com/permissions. This revokes Magister's OAuth grant at the Google level even if the integration is still listed inside Magister.
Webflow Site Content
When you connect Webflow, Magister uses Webflow's official APIs to access the sites you authorize. Depending on your task, we read site names, IDs and domains; page text, titles, descriptions and Open Graph metadata; CMS collection schemas; and CMS item fields, including draft content. We use this content to answer your questions, analyze your site, prepare content, and carry out the changes you request. Publishing and other live Webflow changes require your approval of the specific action.
While a site is connected, on any plan, Magister can also review its CMS collections and items about once a week to keep evidence-backed notes on what has worked for that brand. These reviews cannot publish or change anything.
Where Webflow content goes
- Magister-hosted AI: Webflow page text, CMS fields and other relevant results can become part of prompts, tool results and conversation context sent to the model serving the task, including configured fallback models. Automated audits, suggestions and the periodic reviews above use task-assigned models. The provider can be Anthropic or OpenAI directly, or OpenRouter and its configured inference providers: Google Cloud Vertex AI, Fireworks AI, Together AI, Parasail, or Novita AI. A request follows the task's model and any fallback routes; we do not send every request to every provider. Chat uses these routes in the hosted Agent experience and its free trial. Automated audits, suggestions and periodic reviews use them on every plan, including Connect.
- External AI clients: When you use Connect, or the MCP connector included with Agent, requested Webflow tool results are returned to the external client you authorized, such as Claude or ChatGPT. That client may include the Webflow content in its own AI context and conversation history. Its processing, retention and deletion are governed by your agreement and settings with that provider. Magister does not control those copies.
- Hosting and storage: Vercel hosts our web application, Fly.io runs our Gateway and agent infrastructure, and Supabase stores account, conversation and brand records. Webflow content used in those records or agent workspace files is processed by the corresponding hosting provider. The Sub-Processors page identifies our processing providers.
- Optional SEO research: An audit can send your site domain and keyword or location queries derived from the task to DataForSEO for search and SEO research. This is separate from fetching your page and CMS content through Webflow's API.
- Other services you connect: If you ask the assistant to use another connected integration or MCP server, the content you ask it to pass along goes to that service under that service's terms.
- Error monitoring: When a request fails, the error report sent to Sentry can include the request or message being processed, which can include Webflow content. We remove credentials from these reports, but not other content.
We do not train our own AI models on your Webflow content. Our hosted model routes use the API and routing controls described in Third-Party Services above; those controls do not change the terms of an external AI client you connect yourself. OAuth credentials are kept server-side and are not included in model context or external-client tool responses.
What we store and for how long
Webflow content Magister reads can be saved in chat messages and tool results, site-audit inputs and reports, marketing plans, brand memory, marketing notes and their supporting evidence, drafts, approval records, and workspace files. Supabase stores database and uploaded-file records; Agent workspace files and local conversation history are stored on Fly.io. Disconnecting Webflow leaves these saved copies in place. Uninstalling Magister from the site in Webflow, or removing its access there, deletes them: Magister emails the organization's owner and, about a week later, deletes the brand's plan, audits, chats, workflows, files and agent memory, unless the brand connects Webflow again or a manager chooses to keep the data. Usage records for billing and the provider's ids for what was published stay, without the content. Otherwise they remain until the corresponding records or workspace are deleted.
Connection records contain encrypted credentials and authorized-site metadata while access is active. After a disconnect, Magister keeps the encrypted authorization and its site metadata apart from the connection, so you can reconnect without authorizing again. While disconnected it is used only to check that the authorization is still valid, not to read or change your Webflow content. It is erased when you uninstall Magister from the site in Webflow, remove its access there, or delete the brand. An encrypted record may remain solely to retry provider revocation when a brand is deleted or connects a different authorization. Magister also keeps the host names of the sites a brand has connected, not their content, so it can keep refusing to read those sites' public pages after a disconnect or uninstall. They are deleted with the brand.
How to remove saved content
- Stop using Webflow for a brand: Disconnect Webflow in your brand's Integrations panel. This does not erase previously saved content, and you can reconnect without authorizing again.
- Remove access and saved content: Uninstall Magister from the site in Webflow, or remove its authorization there. Magister erases the authorization, emails the organization's owner, and about a week later deletes what it holds for the brand, as described above.
- Remove a conversation: Deleting a conversation from the Magister sidebar removes its database conversation and messages. It does not erase the Agent's local conversation history, copies saved in reports, brand memory or files, or an external AI client's history.
- Delete an account or request content removal: Account deletion removes account-owned database records and destroys Agent workspaces belonging to organizations created by that account. Records belonging to other members or organizations, including reports they created, can remain; leaving a shared workspace does not delete its content. Uploaded-file cleanup is separate from database deletion and may require follow-up. To request removal of Webflow-derived reports, brand context, files or remaining copies, contact team@magistermarketing.com with the brand and content concerned. Do not include your OAuth token or password.
- Copies held elsewhere: Manage external AI-client history through that client's controls. Changes already saved or published in Webflow remain in Webflow until you edit, unpublish or delete them there. Magister cannot remove them after access is revoked.
Content we delete can remain in infrastructure backups until they age out: database backups are kept for about 7 days and Agent workspace snapshots for 5 days. Error reports age out under Sentry's retention settings. AI-provider and external-client retention is governed by the respective provider's terms and your configuration, separately from Magister's saved records.
Security
- Encryption at rest: All third-party API keys, OAuth tokens, and BYOK keys are encrypted using Fernet symmetric encryption before storage.
- Encryption in transit: All traffic between your browser, our servers, and third-party services is transmitted over HTTPS/TLS.
- Tenant authorization: User-facing database access is restricted through Supabase Row-Level Security (RLS) policies and server-side authorization checks.
- Credential isolation: OAuth tokens and API keys are never stored on your brand's machine. They are injected server-side by our gateway at request time. The only credential on your machine is a scoped, per-user gateway token.
- Isolated workspace infrastructure: Every brand's machine, on the Connect plan, the Agent plan or the free trial, is dedicated and isolated. Workspaces are not shared across customers.
Your Rights
- Access: You can view your data at any time through the app dashboard, including chat history, usage data, and connected integrations.
- Deletion: You can delete your account from account settings or contact us at team@magistermarketing.com for help. Deletion removes account-owned records and Agent workspaces in organizations created by your account. Shared records belonging to other members or organizations can remain, and uploaded-file cleanup may require follow-up. See Webflow Site Content above for the distinction between disconnecting, uninstalling, conversation removal, and removing saved copies. Limited retention required by law may also apply.
- Export: Contact us to request an export of your data.
- Revoke integrations: You can disconnect any third-party service at any time from Integrations. Disconnecting disables new access. For Webflow, disconnecting ends that brand's link only: nothing is revoked at Webflow, and Magister keeps the authorization encrypted, using it only to check that it is still valid, so you can reconnect. To remove Magister's access, uninstall it from the site in Webflow. Already dispatched requests may finish. When a brand is deleted or connects a different authorization, Magister asks Webflow to revoke any authorization no brand can use again. If that cannot be completed, we retain an encrypted cleanup record solely to retry revocation; it is unavailable for site-data API requests and is deleted when cleanup completes.
Data Retention
- Chat history: Database conversations remain until deleted. Agent-local history and copies in other records have separate lifecycles; account deletion does not erase other members' or organizations' records. See the removal controls in Webflow Site Content above.
- Usage data: Stored until you delete your account.
- Integration tokens: Stored encrypted to support active connections. Some providers retain inactive encrypted connection records after disconnect; these are not used for new provider API requests. A disconnected Webflow authorization stays encrypted, used only to check that it is still valid, until you uninstall Magister in Webflow or delete the brand, as described above.
- Analytics we fetch on your behalf: Metrics read from your connected accounts (Google Analytics, Search Console, Google Ads, Stripe, PostHog) are kept for up to 24 months, so year-over-year comparisons work. They are deleted when you disconnect that integration, and when you delete the project or your account.
- MCP authorization and activity: Connector grants and operational activity are retained under the same account, security, billing, and deletion controls as other Magister usage data. Revoking a connector stops new access; deleting your account removes its associated active grants and records subject to limited retention required by law.
- De-identified service metrics: Account deletion may add coarse usage and cancellation categories to a monthly cohort with no account identifiers, free text, exact timestamps, or exact per-account totals. A cohort is available for service improvement only after at least ten separate account deletions contribute, the calendar month closes, and a one-day publication grace passes. Cohorts below that threshold are deleted within 180 days; qualifying aggregate statistics may be retained for trend analysis and cannot be used to restore or identify a deleted account.
- After a paid plan ends: on the Connect and Agent plans alike, each brand's workspace, files and data are kept for 14 days so you can resubscribe, with a daily email to the account owner in the last 7, then the brand's machine and workspace are permanently deleted.
- On account deletion: Workspaces and subscriptions belonging to organizations created by your account are destroyed and cancelled, and account-owned database records are removed. This does not delete other members' or organizations' records, external-client copies, or content published to connected services. Uploaded-file cleanup may require follow-up; see the removal controls above. Limited retention required by law may apply. Account deletion is irreversible.
Children
Magister is not intended for use by anyone under the age of 18. We do not knowingly collect personal information from children. If you believe a child under 18 has provided us with personal data, please contact us and we will delete it.
International Data Processing
Your data is processed and stored in the United States through our infrastructure providers (Supabase, Fly.io, Vercel). By using Magister, you consent to the transfer and processing of your data in the United States.
Changes to This Policy
We may update this privacy policy from time to time. If we make material changes, we will notify you by email or through a notice in the app. The "Last updated" date at the top of this page reflects when the policy was most recently revised.
Contact
For privacy questions, data requests, or concerns, contact us at team@magistermarketing.com.