Skip to content
Legal

Privacy Notice

Effective November 6, 2026

This notice explains what personal information EigenAtom collects through MemScope, how we use and share it, and the choices you have. It is written to match how the service actually works; the security page has more technical detail.

MemScope is in public beta. We'll update this notice as the service matures and will notify customers of material changes.

1. Who we are and what this covers

EigenAtom Consulting LLC ("EigenAtom", "we", "us") builds and runs MemScope, governed organizational memory for AI. This notice covers the marketing site (memscope.ai), the web app, the MCP server that AI clients connect to, and the API.

It doesn't cover the AI clients you connect (such as ChatGPT, Claude or Cursor) or sign-in providers; their own privacy policies apply to them.

2. Our role: your organization controls its memories

MemScope is sold to organizations. For the memories an organization stores, the organization decides what is saved and who may read it, and acts as the controller (or "business"). We process that content on its behalf as a service provider (or "processor"), following its instructions and our Terms of Service.

For account information, billing, security records and our website, we act as the controller.

If you are a member of an organization, questions about the memories it holds about you should go to that organization first.

3. Information we collect

Account information

Your email address, name and password (stored only as an argon2id hash), when you verified your email, the IP address you signed up from and when you last signed in. If you sign in with Google, Microsoft or GitHub, where offered, we store that provider's identifier for you and the email it reported. We use this to create and secure your account.

Organization and membership

Your organization's name, its members and their roles, teams, projects, invitations, and the policy and settings its admins configure. We use this to run the organization and decide access.

Memory content and what we derive from it

MemScope stores only what you or an AI client explicitly send to it (the save_memory tool or the API); it doesn't read your chat history. For each memory we keep its text and every version, who saved it and through which AI client (and an optional agent label and run id), the labels suggested by AI (such as topic and sensitivity), search embeddings, the access policy compiled for it, and links to related memories. This is your organization's content; we process it to provide the Service.

Search activity

When someone recalls or searches memories we record which memories were returned, scores and timing. We keep a one-way hash and the length of the search query, never its text.

AI client connections

When you connect an AI client we store the client's registration details (name, redirect addresses and metadata it publishes) and the connection between it and you. Its access and refresh tokens are stored only as one-way hashes.

Usage, audit and security records

An append-only audit log of actions (who did what, through which AI client, to which memory, from which IP address, and when), which never contains memory content; web sessions with the IP address and browser user agent; usage counts for plan limits; records of AI requests (model, token counts, cost, timing, no content); rate-limit counters; and operational request logs that hold ids, routes, status codes and timings only.

Billing information

Payments are handled by Stripe. We store the Stripe customer and subscription ids, the plan, its status and billing dates, and the number of billed users. We don't receive or store full card numbers.

Email

We send transactional email (verification, password reset, invitations) and keep a record of the recipient, the kind of message, its subject and whether it was delivered.

Website and analytics

Our servers receive standard request information such as IP address and browser type. If you allow analytics, Google Analytics receives the limited data described under Cookies, local storage and analytics.

4. How we use information

  • to provide the Service: store memories, decide who may read them, and return recall results;
  • to create and secure accounts, authenticate people and AI clients, and prevent abuse and fraud;
  • to bill for paid plans and enforce plan limits;
  • to send service and account email, and respond to support requests;
  • to keep the Service running and improve it, using content-free records such as usage counts, error codes and timings;
  • to comply with law and enforce our terms.

We don't sell personal information, we don't use memory content for advertising, and we don't use memory content to train AI models.

6. How AI processing works

MemScope uses AI models at two moments: when a memory is saved and when someone searches. Requests go through OpenRouter, which routes each one to a model provider, and every request asks for zero data retention and no data collection by the provider.

When a memory is saved, its text is sent to a decision model to suggest labels and to an embedding model so search can find it (a correction goes through the same two steps). The decision model also compares a new memory with similar memories the same person can read, to spot duplicates and replacements, and labels sample text tried in the routing preview. Your organization's external AI setting covers all of these: when it is off, they run on local models inside MemScope instead. It does not cover search: the search query, and the top candidate memories to rank, still go to external models, as the next paragraph explains.

When someone searches, the external AI setting does not apply. Recall always sends the search query to the external embedding model, and sends the query together with the text of up to 40 top candidate memories to the external decision model to rank the results. Candidates are only memories the searcher is allowed to read. So turning external AI off does not keep memory content away from external models during search; don't store content in MemScope that must never reach an AI model provider.

MemScope can also be configured with per-provider sensitivity limits that keep memories above a chosen level away from a provider; the default configuration allows every level, so don't rely on them.

AI output is a suggestion. It never decides who can read a memory: a fixed, deterministic policy does, and AI suggestions can only make a memory more private. We don't use your content to train models.

7. Who can see memory content

  • People in your organization: only those the memory's access policy allows. Organization owners and admins manage people and policy, but being an admin doesn't let anyone read a memory outside its audience.
  • AI clients: an AI client sees only what the signed-in person who connected it may read.
  • EigenAtom staff: platform staff see memory content as XXX by default. They can see it only if an owner or admin of your organization grants support access for 1 hour, 24 hours or 7 days, with a reason. The grant can be revoked at any time, and every time staff reveal content it is recorded in an audit log your organization's admins can see. There is no developer bypass.
  • Logs: memory content and search queries are redacted from application logs.

We may also disclose information when required by law or valid legal process, to protect the rights, safety or security of our users, us or others, or as part of a merger, acquisition or sale of assets, subject to this notice. Where the law allows, we will tell the affected organization about a legal demand for its content.

8. Subprocessors

We use these service providers to run MemScope. Each receives only what it needs for its purpose:

  • DigitalOcean: Hosting: the Kubernetes cluster (New York region), the managed PostgreSQL database and DNS. All MemScope data, including memory content, stored in the United States.
  • OpenRouter: Routes AI requests (labels, related-memory checks, search embeddings and recall ranking) to model providers; every request asks for zero data retention and no data collection. Memory text when a memory is saved, unless your organization turns external AI off. On every search, whatever that setting says: the search query, and the query together with the text of up to 40 candidate memories the searcher may read, to rank the results.
  • Stripe: Subscription billing, checkout and the billing portal. Billing contact and payment details. Never memory content.
  • Resend: Transactional email: email verification, password resets and invitations. The recipient's email address and the message. Never memory content.
  • Google Analytics: Website analytics, loaded only after a visitor allows it. Page paths with ids and query strings removed, and a few named events. Never memory content, memory ids, search queries, emails or names.

Sign-in providers (Google, Microsoft, GitHub) are involved only if you choose to sign in with them, where offered. The current list is also on the security page; we will update it when it changes.

9. Cookies, local storage and analytics

Essential cookies

  • ms_session: keeps you signed in to the web app (HttpOnly, 14 days, or until you sign out).
  • ms_oauth_state: protects sign-in with Google, Microsoft or GitHub (HttpOnly, 10 minutes).
  • memscope_theme and memscope_cashew: remember your display preferences (one year).
  • ms_ga_consent: remembers your analytics choice (one year).

The web app also keeps small values in your browser's local storage, such as which organization you last had open and which setup tips you dismissed.

Analytics

This site can use Google Analytics 4 to understand which pages are useful, but only if you allow it. Without your consent no analytics script loads at all, and a browser set to Do Not Track or Global Privacy Control is never asked and never measured.

When you allow analytics cookies, Google Analytics sets its own cookies (such as _ga) and we send page paths with identifiers and query strings removed (a page like /app/memories/<id> is sent only as /app/memories/:id), and a few named events: that a visit started or completed a signup, created an organization, connected an AI client, upgraded a plan, and how many memories your organization holds as a count. We never send memory content, memory ids, search queries, emails or names. IP addresses are anonymized; ads personalization and Google signals are off.

Your choice is kept in the ms_ga_consent cookie for a year, and you can change it anytime with "Cookie preferences" in the site footer or, when you are signed in, at the bottom of the app sidebar (or by clearing that cookie).

10. How long we keep information

  • Memories are kept until your organization deletes them. Ending a trial or cancelling a subscription doesn't delete them.
  • Forget is a soft delete: the memory leaves recall and listings at once, but its content is kept so authorized people can restore it.
  • Permanent purge: an organization owner can permanently purge a memory, or every memory a member wrote. The content of every version, its search embeddings and its AI labels are removed; a record without content (ids, timestamps, audit history) remains.
  • Accounts and organizations are kept while they are active. To delete an account or an organization, contact us at contact@memscope.ai; we will delete or anonymize the data, except what we must keep for legal, security or billing reasons.
  • After an account or organization is closed or terminated: if you ask within 30 days, we will use commercially reasonable efforts to let you export its content; after that period we may delete it. Deletion is not automatic: to have the data deleted, ask us at contact@memscope.ai.
  • Audit records are kept as a security record of what happened. They never contain memory content.
  • Operational request logs hold ids, never memory content, and are deleted after 30 days.
  • Backups: our managed database provider keeps backups on its own schedule, so deleted data can remain in backups until they expire.
  • Billing records are kept as long as tax and accounting law requires.

11. How we protect information

Access to memories is decided by one deterministic policy, with no admin bypass. Each organization's data is kept separate in every lookup. Data is encrypted in transit with TLS and at rest by our managed database provider's disk encryption. Passwords are hashed with argon2id, session tokens are stored only as hashes, AI-client connections use OAuth 2.1 with PKCE, and memory content is redacted from logs.

No system is perfectly secure. If we learn of a breach affecting your personal information, we will notify affected organizations and people as the law requires. More detail is on the security page.

12. International transfers

MemScope is hosted in the United States (DigitalOcean, New York region). If you use it from elsewhere, your information is transferred to and processed in the United States, and our subprocessors may process it in other countries. If your organization needs a data processing agreement or transfer terms, contact us at contact@memscope.ai.

13. Your rights and choices

Depending on where you live, you may have the right to access, correct, delete or receive a copy (portability) of your personal information, to object to or restrict some processing, and to withdraw consent. In the EEA and UK you can also complain to your data protection authority.

US state privacy rights, including California. Residents of California and other US states with privacy laws may have the right to know what personal information we collect and how we use and disclose it, and to access, correct and delete it. We don't sell personal information and we don't share it for cross-context behavioral advertising. We don't use sensitive personal information to infer characteristics about you. We won't discriminate against you for exercising your rights.

In the app, any member can export the memories they are allowed to read as JSON, and the people allowed to can correct or forget memories.

14. How to exercise your rights

For memory content, contact your organization's admin: your organization controls that content, and we will help it respond to requests.

For everything else (your account, billing, our website), email contact@memscope.ai. We may need to verify your identity before acting, and you may use an authorized agent where the law allows. We respond within the time the law requires.

15. Children

MemScope is a business service and is not directed to children under 16. We don't knowingly collect their personal information; if you believe we have, contact us and we will delete it.

16. Changes to this notice

We may update this notice. We will post the new version here with a new effective date, and for material changes we will also notify organization owners by email or in the app.

17. Contact

Questions or requests about this notice: contact@memscope.ai. Security reports: contact@memscope.ai (see the reporting policy).