Memeri Privacy Policy

Effective date: 15 June 2026 Who we are: Memeri Ltd (company number 17051682), 20 Wenlock Road, London, England, N1 7GU ("Memeri", "we", "us") Contact: privacy@memeri.ai · Security: security@memeri.ai

Memeri (memeri.ai) is a persistent AI workspace for software development. You connect your own AI coding agents (such as Claude Code, ChatGPT, or Codex) to Memeri, and Memeri stores your project memory — workflows, jobs, decisions, conversation context, and activity telemetry — so your agents can pick up where they left off.

This policy explains what we collect, why, where it goes, and what your rights are. We've tried to write it the way we'd want to read it: specific, honest, and without padding. Memeri is in private beta. Self-serve export and account deletion are now built (Settings → Data & Privacy); where other data-management details are still maturing, we say so plainly.

The short version

• We store a lot of your project context server-side — including AI conversation transcripts, work logs with git diffs and file paths, decisions, and notes. If you wouldn't put it in Memeri, don't let your agent put it there either.
• Your AI agents run on your machine, under your accounts. We don't sit between you and Anthropic/OpenAI billing. Whatever your agent sends to its AI provider is governed by that provider's terms, not ours.
• No cookies, no trackers, no ads. We use browser localStorage for login and preferences. That's it.
• You can export or delete your account yourself — Settings → Data & Privacy. Account deletion is a true erasure with a 30-day grace period you can cancel; until you delete, your data is kept while your account exists (Section 8).
• We're a small beta product. We list our known limitations in this policy instead of hiding them.

1. Who this policy covers

This policy covers:

• Account holders — developers who sign up for Memeri.
• Visitors — anyone who loads our web app.
• Third parties whose data appears in your projects — for example, people who committed to a git repository you link, or people mentioned in your conversations and notes. See Section 5; this matters and most policies bury it.

For the personal data of account holders and visitors, we act as the data controller. For personal data contained inside the project content you and your agents store with us (including data about third parties), our role is more nuanced: we act as a processor/service-provider for user-supplied content, and as controller for account and telemetry data.

Memeri is intended for users aged 18 or older and is not directed at children.

2. What we collect

Plain-language summary: account details, request logs, detailed telemetry about what your AI agents do, and — the big one — the project content and AI conversations you and your agents store with us.

2.1 Account and identity data

• Email address, username, and a hashed password (bcrypt; we never store or log your password in plain text).
• Email-verification tokens (stored hashed, valid 24 hours, single use).
• Login metadata (last login time, login count) and your settings.
• Connection token — a long-lived credential (currently valid 365 days) your agents use to authenticate to Memeri. Honesty note: this token is currently stored in plain text in our database and returned in some API responses. We're telling you because you should treat it like a password and regenerate it if you suspect exposure. Hardening this is on our roadmap.
• Gateway API keys (stored hashed, with a plaintext prefix for identification) if you connect a REST-based agent (for example, ChatGPT) to the gateway.

2.2 Network, device, and usage data

• IP address and browser/agent user-agent string for every authenticated API request, stored in our usage logs, plus IPs/user-agents in server request logs (which are held on our hosting provider's log infrastructure, see Section 6) and in session metadata.
• Security-event records (IP, user-agent, attempted path) when something trips our abuse detection.
• Honesty note on logs: a few streaming/tunnel endpoints accept your authentication token in the URL query string. We now redact tokens and similar secrets from our own application request logs, but the token still travels in the URL and may appear in our hosting provider's edge/proxy logs (Section 6), on retention we do not directly control. Login and signup events can also write your email address into server logs. Moving tokens out of URLs entirely is on our hardening roadmap; until then, regenerate your connection token if you suspect exposure.
• Feedback reports you send via the in-app "Report issue" widget: your message, your user-agent, recent console errors (which can include file paths or code snippets visible in your session), and whether you're OK with us contacting you.

2.3 AI-agent activity telemetry

Because Memeri's core job is showing you what your agents are doing, we record detailed telemetry about agent activity:

• Per-tool-call events: which tool an agent called, when, success/failure, error messages, truncated parameter and result summaries, and which job/workflow/session it belonged to. We sanitize parameters (secret-pattern redaction and truncation) before storing, but error messages and summaries are free text and can incidentally contain file paths or code fragments.
• Raw tool-call arguments and results for requests routed through our gateway, retained for request de-duplication. Honesty note: unlike the telemetry above, these are currently stored unsanitized and can contain whatever your agent sent — including code, file contents, or secrets pasted into a prompt. Treat agent inputs accordingly.
• Aggregated development-session telemetry: file read/write counts, the primary directories and file paths worked in (local paths often include your OS username), work-phase summaries.
• Agent identity and presence data: which agents are connected, their names, status, heartbeats, role assignments, and behavioral scoring (e.g., protocol-skip events).

2.4 Project content and "memory" (the most sensitive category)

This is everything you and your agents deliberately put into Memeri, and it can contain anything — code, architecture, business plans, names, credentials you shouldn't have pasted:

• AI conversation transcripts. When conversation sync is active, the individual messages you exchange with your agent (your prompts and the agent's replies) are uploaded from your machine and stored server-side, along with semantic embeddings of them. The primary copy of console/session transcripts lives on your machine (e.g., ~/.memeri-console/, session .jsonl files), but the server-side copy exists and this policy covers it. If you tell your agent something, assume Memeri stores it.
• Projects (name, description, linked repo URL, and the workspace file path, which can reveal usernames).
• Workflows, jobs, roadmaps, todos, specs, documents, and planning content.
• Job work logs including evidence: summaries plus captured git status / git diff --stat / git log output, files changed, and commands run.
• Artifacts (decisions, checkpoints, notes), intelligence items (decisions/learnings with reasoning and evidence), scratchpad notes, agent handoffs, in-app chat threads and attachments, discussion threads.
• AI task-queue prompts and tool calls awaiting your approval (including their arguments).
• Semantic embeddings of the above, computed by an embedding model running on our own servers — no third party receives your text for embedding.

2.5 Data from linked git repositories

If you link a repository, we poll it (currently public repos, via the GitHub API) and ingest branch and commit metadata — including commit author names and email addresses, which may belong to people who have never used Memeri. See Section 5.

2.6 Terminal and local-machine data

Your agents execute commands on your own machine via the Memeri local console. Terminal command history is held in server memory during a session but is not persisted to our database unless an agent explicitly posts output into a work log or artifact. Console transcripts, scrollback, and workspace layouts live in ~/.memeri-console/ on your machine, not on our servers. Setup also installs a small session hook into your local Claude settings so conversations can be linked to jobs — we're disclosing this because it modifies a config file on your machine.

2.7 What we do NOT collect

• No cookies. No analytics or tracking scripts of any kind (no Google Analytics, no session replay, no pixels). We verified this against our actual client code.
• No payment data — payments are not live yet. When we add a payment provider, we will update this policy before launch (naming the provider and its role, including any merchant-of-record relationship), including a fresh cookie assessment, because payment scripts typically set cookies.
• No precise location, no contacts, no device sensors.

3. How we use your data (and our lawful bases)

Plain-language summary: we use your data to run the product, keep it secure, and email you about your account. We don't sell it, we don't advertise with it, and we don't use it to train AI models.

Things we do not do:

• We do not sell or share your personal data for advertising (see Section 11 for the CCPA version of this statement).
• We do not use your content or conversations to train AI models, ours or anyone else's.
• We do not send marketing email. The only email we send today is transactional (verification, account notices). If we ever add product newsletters, they will be opt-in.
• We do not make automated decisions about you that produce legal or similarly significant effects (GDPR Art. 22). The agent "behavioral scoring" described in Section 2.3 is telemetry about your AI agents' protocol adherence, shown to you in the product; it is not profiling of you and is not used to make decisions affecting your account or access.
• As a matter of design policy, Memeri does not send your content to large-language-model providers on our own account for routine features — your own agents do the AI work under your own provider accounts.

4. Your AI agents and their providers (read this section)

Plain-language summary: you bring your own AI. Whatever your agent sees, its provider sees — under that provider's terms, not ours.

Memeri is a workspace around AI agents you supply. You authenticate to Anthropic, OpenAI, or another provider directly; we never own that relationship, resell model output, or sit between you and the vendor's billing.

This has a privacy consequence you should understand clearly:

• When your agent calls a Memeri tool, the result of that tool — file contents, git diffs, project memory, intelligence items, terminal output — is returned into your agent's context and therefore transmitted to your AI provider (e.g., Anthropic or OpenAI).
• That data is then subject to your provider's privacy policy, retention rules, and training policies — not this policy. Check your provider's settings (e.g., training opt-outs) if that matters to you.
• Your AI providers are not our sub-processors for this data flow; they are services you chose and contracted with.

Separately: your agents execute real commands on your machine with your privileges. Memeri hosts the surface, not the compute. What an agent reads from your disk can end up in Memeri (via work logs, conversation sync) and at your provider (via the agent's context).

5. Data about other people (third parties)

Plain-language summary: if you link a repo or talk about people to your agent, data about those people ends up in Memeri. Make sure you're allowed to put it there.

Two ways data about people other than you enters Memeri:

1. Linked repositories. We ingest commit author names and email addresses from repos you link — including contributors who have never used Memeri. We use this only to show you your project's commit activity. If you are a commit author whose data appears in someone's Memeri project and you want it removed, contact privacy@memeri.ai.
2. Your content. Conversations, notes, and code you store may mention or include other people's data. You are responsible for having a lawful basis to put third-party personal data into Memeri. Where we store such data on your behalf, we act on your instructions.

6. Who we share data with (sub-processors and recipients)

Plain-language summary: a hosting provider sees everything (that's how hosting works), an email provider sees your email address, GitHub is involved if you link a repo, and one CDN currently sees your IP on terminal pages. No data brokers, no ad networks.

6.1 Service providers

6.2 Planned (not yet active)

• A third-party payment processor (payments) — not integrated yet. The specific provider will be named here, with its role (including any merchant-of-record relationship), before payments launch.

We currently have no analytics, error-tracking, or advertising vendors. If we ever add one, we will update this policy first.

6.3 No platform-side AI providers

Memeri does not send your content to large-language-model providers on our own account. All AI work runs on your own connected agents under your own provider accounts (see Section 4). We do not use platform API keys to send your chat text, project context, or workflow prompts to OpenAI, Anthropic, or any other LLM provider, and accordingly we do not list any such provider as a Memeri sub-processor.

6.4 The connection tunnel

When your agents talk to your machine through Memeri, traffic passes through a WebSocket relay hosted on our infrastructure. This relay forwards file contents, terminal commands and output, and git diffs between your machine and your agents. This traffic is encrypted in transit (TLS) but is not end-to-end encrypted — it transits our servers in a form our infrastructure could technically observe. Additionally, certain file operations fall back to our server's filesystem when no tunnel to your machine exists; file operations are not guaranteed to be local-only.

6.5 Other disclosures

We may disclose data if required by law or legal process, to protect the rights, safety, or property of Memeri or others, or as part of a merger, acquisition, or asset sale (in which case this policy continues to apply and we'll notify you).

7. Browser storage (the "cookie" section)

Plain-language summary: zero cookies, zero trackers. We use localStorage for your login token and preferences.

We set no cookies and run no tracking or analytics scripts. We use browser storage as follows:

• localStorage: your login token (7-day) and refresh token (30-day), gateway connection settings (including, currently, a gateway API key in plain text — see honesty note below), UI preferences, draft workflows, cached activity feeds, and recent searches.
• sessionStorage: navigation state and short-lived UX flags, cleared when the tab closes.

Honesty notes: (1) logging out clears your auth tokens but currently does not clear drafts, cached feeds, or gateway settings — on a shared computer, use a private window or clear site data manually; (2) if you connected a REST-based agent via a gateway API key, that key sits in localStorage in plain text. Both are on our hardening roadmap.

Under the EU ePrivacy rules, this storage is either strictly necessary (login) or functionality you asked for (preferences, drafts), so no consent banner is required. We will reassess this the moment we add payments or any analytics.

8. Data retention

Plain-language summary: while your account exists, most data is kept until you delete it — we don't yet run fixed retention schedules for most categories. You can erase your account yourself (Settings → Data & Privacy): a true erasure that completes after a 30-day grace period.

We are a private beta and are still building out automated, category-by-category retention schedules. The current reality:

• Project content, conversations, telemetry, and usage logs are retained for as long as your account exists. When you delete your account (Section 9), they are deactivated immediately and permanently erased after the 30-day grace window — across the platform and the connected gateway, in a single transaction, with your gateway credentials revoked.
• Security and audit logs are an exception: on erasure they are anonymized (your account link removed) rather than deleted, and retained for up to 90 days for intrusion-detection, abuse-prevention, and accountability (legitimate interest / legal obligation), then automatically deleted. They may still contain technical data such as IP addresses for that window.
• The exceptions with working automated cleanup today: streaming output chunks (deleted after ~1 hour) and agent pairing codes (10-minute expiry). Email-verification tokens expire after 24 hours.
• Backups: we keep nightly database backups for approximately 14 days. Data deleted from the live database can persist in backups for up to that period, and a backup restore could temporarily resurrect deleted data — if that happens after we've processed your erasure request, we will re-delete it.
• Deleting individual items in-app is often a soft-delete today. Deleting some content types (projects, specs, intelligence items, scratchpad notes, attachments) currently archives or deactivates the data rather than erasing it, even where the interface says "deleted". Residual data from earlier product versions also persists in legacy storage. Account-level erasure (Section 9) is a true, irreversible delete; for individual items, treat in-app deletion as "hide" until per-item hard-delete ships, and use the account erasure when you need that data actually gone.
• Server logs are retained per our hosting provider's log retention, which we do not directly control.

As we ship data-lifecycle tooling, we will publish concrete retention periods here. Until then, the honest statement is: assume indefinite retention unless you ask us to delete.

9. Your rights (GDPR / UK GDPR and generally)

Plain-language summary: you have real legal rights, and we'll honor them — but today most of them are fulfilled manually by a human, not a button. Email privacy@memeri.ai.

If you're in the EU/UK (and as our default for everyone), you have the right to access, correct, export, delete, and restrict or object to processing of your personal data, and to withdraw consent where consent is the basis. You also have the right to complain to your supervisory authority.

How this works in practice today:

• Access / export: Settings → Data & Privacy → Export gives you a full, machine-readable export of your data — projects, conversations, memory, telemetry, and profile — with credentials redacted from the file. You can also email privacy@memeri.ai and we'll compile and deliver it within 30 days (usually much faster at our current scale).
• Deletion: Settings → Data & Privacy → Delete account schedules a true erasure. Your account is deactivated immediately and permanently deleted from live systems after a 30-day grace period — we email you a cancel link so you can reverse it any time within that window. The erasure also revokes your connected gateway credentials. Backup copies age out within ~14 days after erasure; security/audit logs are retained anonymized (Section 8). You can also email privacy@memeri.ai from your account email. We'll confirm when done.
• Correction: you can edit most content in-app. One design caveat: historical activity records are immutable by design — renaming an agent or project updates its live identity but does not rewrite the names embedded in past event logs. If a correction matters to you and the in-app tools can't do it, email us.
• Objection / restriction: email privacy@memeri.ai and we'll work it out, including stopping repo polling or feedback contact.

We will verify requests (normally by matching your account email) and will not discriminate against you for exercising any right.

10. International transfers

Our infrastructure is hosted on Railway. If you are in the EU, UK, or elsewhere outside that region, your data is transferred to and processed there.

11. California (CCPA/CPRA)

For California residents: we collect the categories described in Section 2 (identifiers, internet/network activity, and user-provided content), for the purposes in Section 3. We do not sell personal information and we do not share it for cross-context behavioral advertising, and we have not done so in the preceding 12 months. We use no third-party advertising or analytics, so there is nothing for a "Do Not Sell Or Share" link to switch off; for the same reason, opt-out preference signals such as Global Privacy Control have no selling/sharing to opt out of (we set no cookies for such signals to govern). The rights to know, delete, correct, and non-discrimination apply, via privacy@memeri.ai, on the same manual process described in Section 9.

12. Security

Plain-language summary: real measures exist, real gaps exist, and we're a beta — don't store anything in Memeri whose leak you couldn't tolerate.

Measures in place: passwords hashed with bcrypt and never logged or returned; email-verification tokens stored hashed with short expiry; parameterized SQL throughout; HMAC-verified webhooks with timing-safe comparison; TLS in transit; provenance logging on agent actions; secret-pattern redaction in telemetry (with the coverage limits described honestly in Section 2.3 — redaction is not exhaustive); a verified backup-restore drill; and a 2026 security audit after which all 17 identified authentication issues were fixed and verified in production.

We are deliberately not going to promise "bank-grade security." Memeri is a private beta and we maintain a known-issues backlog (including credential-handling hardening noted in Sections 2.1 and 7). No system is perfectly secure, and we make no absolute guarantee. Practical advice we actually mean: don't paste production secrets into agent conversations, and treat your connection token like a password.

If we learn of a personal-data breach, we will assess it and notify affected users and regulators as required by applicable law (including GDPR's 72-hour authority-notification rule where it applies), without undue delay.

Found a vulnerability? Please tell us: security@memeri.ai. We commit to responding and to not pursuing good-faith researchers.

13. Changes to this policy

We'll update this policy as the product evolves — notably before launching payments, adding any analytics or error-tracking vendor, adding any platform-side AI provider, or changing hosting. Material changes will be announced by email or in-app notice with the new effective date. Continued use after the effective date means the updated policy applies.

14. Contact

• Privacy requests and questions: privacy@memeri.ai
• Security reports: security@memeri.ai
• Postal: Memeri Ltd, 20 Wenlock Road, London, England, N1 7GU

Last updated: 18 June 2026