Skip to content

Privacy Policy

Last updated: 2026-03-12

This policy describes what data agentgovernance.dev collects, how it is stored, and how it is used. It covers both the documentation site and the governance API.

About this site

agentgovernance.dev is a hosted governance API for AI agent operations, maintained by an independent researcher. Questions about this policy can be directed to the GitHub repository: github.com/agentgovernance/governance-engine.

What is collected

1. Site analytics

When you visit agentgovernance.dev or agentgovernance.dev/docs, two analytics systems record your visit:

Plausible Analytics — a privacy-first analytics tool. Plausible does not use cookies, does not store personally identifiable information, and does not track users across sessions or sites. It records: page URL, referrer, browser family, device type, country (derived from IP, IP not stored). Plausible's privacy policy: plausible.io/privacy.

Google Analytics — records standard web analytics data including anonymized IP, pages viewed, session duration, referrer, browser, and device. Google Analytics uses cookies. Google's privacy policy: policies.google.com/privacy. You can opt out using the Google Analytics Opt-out Browser Add-on.

2. API registration

When you register an API key at /api/v1/register, the following is stored:

  • Email address — stored in plaintext. Used to identify your account and for future account management communications.
  • Organization name and slug — stored in plaintext.
  • API key — generated by the server, returned to you once, and immediately hashed (bcrypt). The plaintext key is never stored. Lost keys cannot be recovered.

3. API usage — anonymous evaluate endpoint

When you call POST /api/v1/evaluate without an API key, the following is recorded:

  • Hashed IP address — your IP is hashed (SHA-256) before storage. The raw IP is never written to the database.
  • Input character count — the number of characters in your submitted text. The text itself is not stored.
  • Input mode — whether you submitted plain text or an agent trace format.
  • Input framework — the detected framework label (langgraph, autogen, openai, unknown), if detectable.
  • Evaluation result — the full pattern detection result JSON, including which patterns were detected, confidence scores, and severity ratings.
  • Timestamp and expiry — evaluation records expire and are deleted 30 days after creation.

What is not stored from anonymous evaluate calls: The text you submitted. It is analyzed in-memory and discarded.

4. API usage — authenticated observe endpoint

When you call POST /api/v1/observe/assess with an API key, the following is recorded:

  • Organization ID, Agent ID, Run ID — identifiers you provide to associate the observation with a governance run.
  • Pattern ID — which behavioral pattern was detected (e.g., BP-001).
  • Confidence and severity — the detection result scores.
  • Evidence — the first 500 characters of your submitted text (only for text where a pattern was detected). Text that does not match any pattern is not stored.
  • Detection method — how the detection occurred (regex, semantic, context-aware).
  • Audit trail entry — a record that patterns were detected, including which pattern IDs and the agent/run context. The text content is not duplicated in the audit trail.

What is not stored from observe calls: Text content that produces no detections. Text beyond the first 500 characters, even for positive detections.

5. Governance lifecycle data

When you use the governance gate endpoints (/api/v1/govern), the following is recorded:

  • Gate name, gate action (submitted, approved, rejected), actor identifier, and timestamp.
  • Evidence metadata you include in the gate submission (e.g., commit references, test result summaries). This content is stored as-is and is not validated or processed.

No agent output text is stored via the govern endpoints.

6. Rate limit tracking

To enforce daily rate limits, a rate limit bucket is maintained keyed on:

  • Anonymous requests: SHA-256 hash of your IP address + date.
  • Authenticated requests: your API key ID + date.

Raw IP addresses are never stored for rate limit purposes.

How data is stored

All data is stored in a PostgreSQL database hosted on Neon (neon.tech). Neon is a serverless Postgres provider. Data is stored in the US region. Neon's privacy policy: neon.tech/privacy.

The API server runs on Replit infrastructure. Replit's privacy policy: replit.com/privacy.

Data retention

Data type Retention period
Anonymous evaluation results 30 days (auto-deleted via expiresAt)
Authenticated evaluation results 30 days (auto-deleted via expiresAt)
Behavioral observations (authenticated) Duration of organization subscription
Gate events and governance runs Duration of organization subscription
Audit trail Duration of organization subscription
Rate limit buckets Expire at midnight UTC daily
API key records (hashed) Until key is revoked or organization is deleted
Email addresses Until account deletion is requested
Plausible analytics Per Plausible's retention policy
Google Analytics Per Google's retention policy (default 26 months)

Data isolation

All authenticated API data (observations, runs, gates, agents, audit trail) is partitioned by orgId. Your data is never visible to other organizations. There are no shared tables or shared result sets between organizations.

Commitments

  • No model training. Agent output text submitted to the governance API is used only to produce detection results. It is not retained for model training, benchmarking, or research.
  • No data sales. Data is not sold, licensed, or otherwise distributed to third parties.
  • No plaintext API keys. API keys are hashed immediately on generation. A copy of a lost key cannot be provided.
  • No raw IP storage. All IP-based records use a one-way hash.

Your rights

You may request:

  • Access — a copy of data held associated with your email or organization.
  • Deletion — removal of your organization, all associated agents, runs, observations, and your email record.
  • Correction — correction of your organization name or email if incorrectly recorded.

To make a request, open an issue at github.com/agentgovernance/governance-engine or use the contact form.

Changes to this policy

Material changes to this policy will be noted in the repository commit history with a dated update to this document. The date at the top of this page reflects the last revision.