Privacy
Effective 02 Aug 2026 · applies to the MAESTRO agent and try-maestro.pro
What the agent sends us (telemetry)
- Execution events: recipe name, step name, start/success/fail, agent version.
- System context: OS release string and platform (e.g. "Ubuntu 24.04, x86_64"). No hostnames.
- Error tails on failure: the last lines of the failing command's output, redacted on your server before sending — passwords, tokens, keys and long encoded strings are stripped by the agent itself. You can read the redaction code: it's in the single agent file.
- Server fingerprint: a one-way hash of your machine-id — we can count "same server", we cannot identify it.
- IP addresses are not stored — only a salted one-way hash, for abuse protection.
Why we collect it
- Healing: the failure context is what the AI uses to propose your fix.
- Learning: repeated failures harden the recipes, so the same thing never breaks twice — for anyone. Successful fixes are cached by an anonymous error signature.
- License accounting: the server fingerprint counts servers per key. Nothing else.
What we never do
- No sale or sharing of data with third parties (AI processing runs via Anthropic as processor).
- No collection of your users' data, file contents, databases or mail.
- No hidden telemetry: the agent prints a notice on every install, and --no-telemetry (or MAESTRO_NO_TELEMETRY=1) turns it off entirely. Healing requires the failure context to function — that part is inherent to the feature you invoke.
Retention & your rights
Telemetry events are kept up to 12 months, then aggregated or deleted. To exercise access/erasure rights or ask anything: hi@pentimento.pro. Supervisory authority for EU residents: your local DPA. Operator: PENTIMENTO (pentimento.pro), Bulgaria.