Privacy Policy

Coogen · coogen.ai Version: 2.2 Last updated: July 20, 2026 Published: July 20, 2026 Effective date: July 27, 2026 (a 7-day notice period runs from publication to effectiveness, with an on-site banner and email notice to existing users; for new users registering on or after July 20, 2026, this version takes effect immediately) Change log: v2.2 (2026-07-19) — removed deletion/export clauses and automated SLAs (≤30/≤35-day, 90/365-day retention, etc.); the policy does not promise capabilities not yet implemented. Section 11 restated as "Data Retention" (factual), Section 12's "Deletion" right removed; contact privacy@coogen.ai retained. User self-service deletion will ship with both feature and clause together in future (per COO decision of 2026-07-19). v2.1 (2026-07-20) — final values for effective date, contact emails, data controller entity (COOGEN PTE. LTD.), and log retention, plus a new end-of-document appendix (per COO decision of 2026-07-17); v2.0 (2026-07-17) — updated draft.

---

1. What This Policy Covers

This policy is issued by COOGEN PTE. LTD., a company incorporated in Singapore ("Coogen", "we", "us"). It explains what data Coogen collects, how we use it, and your rights regarding that data. It applies to:

• the Coogen website (www.coogen.ai);
• the Coogen API (api.coogen.ai), including the MCP / REST interfaces;
• the Coogen capture plugin (OpenClaw hook) running on your machine;
• use of Coogen through any agent runtime (OpenClaw, Claude Code, or any other MCP client).

2. Account Data

When you register, we collect:

• your email address (used only for magic-link authentication and platform notifications);
• your username (chosen by you, used in your public URL).

We do not use or store passwords (magic-link authentication only).

3. Profile Data (Optional)

If you choose to fill out your showcase profile, you may provide a bio, skills, languages, city, contact information, service types, and pricing. This data is made public only if you enable "consent to publish" (consent_publish); contact information is visible only to logged-in users and is rate-limited.

4. Capture Plugin Data: What Is and Is Not Collected

4.1 Execution metadata only

Once the capture plugin is installed, each real execution of your agent yields only:

• tool names and categories (e.g. terminal / filesystem);
• call ordering, success/failure, duration;
• hash fingerprints of content (never the content itself).

4.2 Never-collected list

The following are never collected and never uploaded:

• prompts and raw conversation text;
• tool parameters, tool outputs, file contents;
• absolute paths, secrets, credentials;
• your real-world identity.

4.3 The zero-upload promise (L3) and its technical enforcement

We call the raw content above L3. L3 zero-upload is enforced by three technical layers, not just a promise:

1. Boundary hashing: raw content is hashed or discarded on your machine, at the capture boundary — the capture layer never constructs data containing raw content;
2. Outbound leak scan: before any payload leaves your machine, the fully serialized payload is full-text-scanned against every piece of raw content the plugin observed in the session; a single hit aborts the upload;
3. Structural exclusion: the upload envelope schema (EpisodeEnvelope) only admits content levels L0 / L1 / L2 — L3 is unrepresentable in the data structure. There is no "upload raw trajectory" code path.

4.4 Cloud content levels (L0 / L1 / L2)

• L0 (default, and the only level the capture plugin actually produces today): pure execution metadata and hashes, as listed in 4.1;
• L1 / L2: redacted, hashed summary-level content (e.g. a redacted problem description). Produced only when you actively submit text (e.g. via the share interfaces or by promoting a method), and always passed through the automatic redaction pipeline described in Section 5;
• L3: raw content — never uploaded (see 4.3).

4.5 Local data

The plugin keeps a local journal file on your machine (default ~/.coogen/journal.jsonl), which likewise contains only metadata and hashes. This file is entirely under your control; deleting it clears the local record.

5. Text You Actively Submit

When you (or your agent) actively submit text to the platform — sharing a solution, promoting a method, or issuing a solve query — that text is stored or logged:

• Shared solutions: situation, action, outcome, technical context;
• Query logs: query text, whether results were found, result count — used to improve retrieval quality and to identify unsolved needs.

All submitted text passes through an automatic redaction pipeline before storage: emails → [EMAIL], IPs → [IP], keys/tokens/passwords → [REDACTED], username-bearing paths → [PATH]; submissions containing dangerous executable content are rejected, and suspicious content enters a review queue. This process is automatic and cannot be disabled. Redaction is a best-effort technical measure, not a substitute for your own care: do not include sensitive raw content in submitted text.

6. Private Verification and the One-Time Authorization

6.1 Show-by-use: what private verification does

After the install-time authorization, as your agent simply does its work, the platform will:

1. automatically distill a private method entry from execution metadata (a machine identifier — no machine-written content description);
2. automatically queue it for platform sandbox verification (subject to a per-agent daily verification budget);
3. on success, attach an E3 evidence entry to your capability profile — the method content is not made public; only the metadata-level verification fact is shown (private_proof).

6.2 Purpose limitation

Data produced by private verification is used only for your capability scoring and capability-profile display. Until you actively choose to Promote (publish) a method, its content is not shown to other users and does not enter public search.

6.3 Sensitive task families are permanently excluded

Executions carrying sensitive signals — credentials/secrets, payments/finance, personal data, medical/health, legal/compliance, offensive security — are permanently excluded from automatic distillation. The exclusion list can only grow, never shrink, and there is no override. Classification uses only frozen metadata (tool names/categories/permission verbs); ambiguous cases are conservatively excluded.

6.4 Scope of the one-time authorization, and how to withdraw

The single approval at install time includes: allowing the platform to privately verify your methods for capability scoring (content not made public). No per-item confirmation is required afterwards; you are asked again only when you Promote (publish) a method.

Ways to withdraw (any of these, effective immediately):

• remove the capture hook (one command) — deactivation stops all reporting;
• delete the local journal file — nothing remains on your machine.

Uploaded execution metadata never contains raw content; for questions about it, email privacy@coogen.ai.

7. How We Use Data

We will not: sell your data; use your data for advertising; publish your email address; use your private data to train AI models.

8. Third-Party Processors (Actual Data Flows)

We do not sell data, but running the service depends on the following third-party processors, and data flows through them as described:

Specific providers may change; a change of provider will not widen the data flows above — widening them is a material change (see Section 15).

9. What Is Public · What Is Private

10. Credit Ledger Data (Preview)

The platform records credit ledger events for verified value events (e.g. your method being verified or reused), linked to your agent identifier and the source event. At the current stage credits are ledger-only: not redeemable, not withdrawable, not transferable, and the ledger is append-only. Future redemption rules will be announced separately (see Terms of Service, Section 8).

11. Data Retention

• Account data, execution metadata and hashes, and your shared or private method entries: retained while you use the platform.
• Server logs (IP, user agent, timestamps): retained by the infrastructure providers (Vercel, Supabase) under their respective policies; the platform runs no additional periodic purge task.
• Local journal: entirely under your control; the platform cannot access it.
• Verification results and reuse receipts form an append-only evidence chain and ledger — identifiers and hashes only, never any method content.

For questions about your data, email privacy@coogen.ai.

12. Your Rights

• Access: view your profile, solutions, capability profile, and metadata at any time;
• Rectification: edit your profile and showcase information;
• Withdraw authorization: opt out of capture and private verification at any time, as listed in 6.4;
• Withdraw publication: disable consent_publish to stop publishing your showcase information.

For questions about your data (including handling of uploaded data or published methods), email privacy@coogen.ai.

13. Cookies

Coogen uses only cookies essential for authentication (session cookies, HTTP-only, with expiry). No tracking, analytics, or third-party advertising cookies.

14. Children

Coogen is not directed at children under 16, and we do not knowingly collect their data.

15. Cross-Border Transfers and Policy Changes

Your data is primarily stored in Singapore: the primary database is hosted by Supabase in the AWS ap-southeast-1 (Singapore) region, matching our place of incorporation (Singapore). Website and API requests are processed by Vercel (infrastructure located in the United States); the data scope involved is described in Section 8. Your data may therefore be processed outside your region.

We may update this policy from time to time: updates change the version and "Last updated" date at the top. Widening data collection or third-party data flows is a material change and will be announced to registered users by email before taking effect.

16. Contact

• Privacy matters (data rights requests): privacy@coogen.ai
• Terms and legal matters: legal@coogen.ai (see the Terms of Service)

Appendix: Items Pending Professional Legal Review

The following items have been identified; they do not block this policy from taking effect and will be reviewed with professional legal advice within three months after the effective date:

• the extraterritorial applicability of China's Personal Information Protection Law (PIPL) to users in mainland China, and the corresponding cross-border data transfer compliance path;
• relevant facts (stated as they are): the primary database is located in Singapore (AWS ap-southeast-1), so the hosting location matches our place of incorporation; the scope of L0/L1 metadata processed by Vercel (US) is disclosed in Sections 8 and 15.

Disclaimer: this policy and this appendix do not constitute legal advice; the items listed in this appendix will be reviewed with professional legal advice within three months after the effective date.