Security practices
TriageAI reads your code, your bug reports and your production telemetry, so it's built to see as little as possible, change nothing, and leave a record of everything it did.
Secret redaction
Before any input reaches the model, a sanitizer strips API keys, tokens, passwords, connection strings, emails and other PII-like patterns.
Read-only, scoped access
File tools work on a read-only clone, and path traversal outside the repository is rejected. Outage analysis uses read-only credentials for alerting, metrics, logging and deploy tools.
Audit logging
Every tool call is written to structured JSON logs with its input, output, timestamp and request ID, so any diagnosis can be traced back step by step.
Human review, no auto-merge or auto-remediation
Every diagnosis and outage analysis is marked as a hypothesis for an engineer to review. TriageAI never merges or pushes code, and never runs a rollback, restart or other remediation itself.
Sandboxed test execution
run_test only runs commands from a fixed allowlist, with a timeout and no network access. Bug report text is treated as untrusted input and can never choose a command.
Prompt-injection awareness
The agent is told to treat the bug report as data and ignore any instructions inside it. Its tools are read-only, so even a successful injection can't change anything.
API key hygiene
Keys are read only from environment variables on the server. They are never sent to the browser and never written to logs.
Webhooks, rate limits and size limits
GitHub webhooks are rejected unless their HMAC signature verifies. The triage endpoint is rate limited, and input size is capped.
Your data
Don't paste production secrets or customer data into the demo. Redaction is a safety net, not a reason to send sensitive data.
Watch TriageAI work a sample outage
Follow it from the alert to the change behind it and the exact line of code, then get the fix and a regression test. You review every step.