Privacy Policy & Telemetry Notice

1. Architectural Foundation: Local Compute & Direct Provider Inference

Mrrlin is engineered around a local-first computing model. When you use Mrrlin with coding agents (Claude Code, OpenAI Codex, or Google Antigravity), all operations conform to our Terms of Service:

• Local Tool Execution: Agent processes execute locally on your hardware as child processes. File modifications, tests, and git operations execute in your workspace directory (cwd) under your local OS privileges.
• Direct Provider Model Inference: When agents reason or generate code, your local CLI communicates directly with your chosen AI provider APIs (Anthropic, OpenAI, or Google) using your local BYOK credentials. These model inference calls do not pass through Mrrlin cloud servers.
• Local Key Storage: Model API keys are kept in your local configuration files or local environment and are never transmitted to Mrrlin cloud servers.

2. Separation of Operational Telemetry and Hosted Collaboration Data

To maintain clarity and transparency, Mrrlin strictly separates machine telemetry from hosted workspace collaboration data:

3. What We Collect in Routine Telemetry

To diagnose operational health, connection reliability, and framework performance, the local bridge transmits aggregated daily operational batches. As defined in our canonical telemetry schema, these metrics consist solely of:

• Tool Execution Sequences: Generic tool names called (e.g., bash, read_file), execution duration in milliseconds, and exit status (success or error).
• Turn-Level Token Metrics: Integer counters representing token volume (inputTokens, cachedInputTokens, outputTokens, and prompt byte size). These are raw counts only; no token text or prompt content is stored or transmitted.
• Plumbing and Health Counters: Monotonic counters for connection lifecycle events (WebSocket disconnects, scheduler ticks, buffer queue statistics, and rate limit flags).
• Gate Consensus Performance: Anonymous consensus records (gate kind, round number, model family identifier, and elapsed duration).
• Hashed Project Partitions: Project slugs used as local partition keys are converted into salted, one-way cryptographic hashes upon server ingest.
• Anonymous Installation ID: An opaque, randomly generated identifier persisted locally on your machine to distinguish multi-machine environments.

4. Two-Tier Protection: Deterministic Redaction & Sandboxed PII Gating

For CLI diagnostic bundles (mrrlin report extract), Mrrlin employs a mandatory, two-stage sanitization pipeline implemented directly in the local CLI:

5. Voluntary Issue Reporting & Diagnostics

You may optionally submit diagnostics to help troubleshoot failures through two distinct channels:

• CLI Diagnostic Bundles (mrrlin report extract): When generating an evidence bundle locally, execution steps, exit codes, and tool excerpts pass through both Tier 1 deterministic regex secret scrubbing (redact.ts) and Tier 2 sandboxed PII model gating (pii-gate.ts) prior to local export or transmission.
• In-App Issue Reports (“Report Issue” Modal): When submitting an in-app report, your entered description and active transcript are transmitted to our coordination API where server-side secret scrubbing redacts known auth tokens and private key headers. Optional user-attached screenshots are stored as binary artifact files and are not inspected by automated text PII filters.
• Full Operator Visibility: In both channels, reports are strictly opt-in and triggered only by explicit operator action. You can inspect all attached context or cancel submission at any time.

6. Account and Authentication Data

To provide access to the collaborative cloud workspace, we collect and store:

• Identity Profiles: User identifier, name, email address, and avatar image provided via GitHub OAuth or Firebase Authentication.
• Session Tokens: Secure HTTP-only cookies used to maintain your authenticated web session.
• Billing Details: Subscription tier and customer identifier managed by Stripe. Mrrlin never stores credit card or bank account numbers.

7. Data Retention and Local Pruning

Local telemetry files stored on your machine in the state directory are partitioned by day and are automatically pruned by our background retention cleaner (pruner.ts) in accordance with local storage limits. Cloud-ingested telemetry rows are retained for operational analysis and automatically aged out after our operational window.

8. Legal Bases for Processing and Regulatory Roles (GDPR & CCPA/CPRA)

Under the General Data Protection Regulation (GDPR) and California Consumer Privacy Act (CCPA/CPRA), we process your personal data under the following legal bases:

• Performance of a Contract (GDPR Art. 6(1)(b)): Maintaining authenticated user accounts, synchronizing cloud workspace items (tasks, plans, wiki), and processing subscriptions.
• Legitimate Interests (GDPR Art. 6(1)(f)): Collecting aggregated operational telemetry counters, monitoring connection health, and preventing fraudulent API usage.
• Consent (GDPR Art. 6(1)(a)): Voluntary submission of diagnostic issue reports.

Mrrlin acts as an independent Data Controller with respect to your account, billing profile, and routine operational telemetry. For project workspace content created and managed by team members within collaborative workspaces, Mrrlin acts as a Data Processor.

9. Your Rights and Data Choices

Depending on your jurisdiction (including rights under GDPR and CCPA/CPRA), you have the right to:

• Access the personal account data associated with your identity.
• Request rectification or deletion of your account and associated profile records.
• Export your workspace items or request data portability.
• Disable optional crash and evidence uploads.

10. Privacy Inquiries and Data Protection Officer

For any privacy requests, questions, or data protection officer inquiries, please contact: