Traccia gives you a clear choice — rich Full-trace mode for debugging, or Metadata-only mode for sensitive production workloads. Capture exactly what your compliance posture allows, no more.
Best for development & debugging. Capture deep execution details:
Best for regulated production workloads. No prompt or completion text.
skip_args / skip_result, and PII redaction — all in the SDK, before any data leaves your process.Redaction is on by default to protect sensitive data out-of-the-box. Enable init(redact_pii=True) to explicitly configure and mask emails, US phone numbers, and SSN-like patterns on sensitive span attributes before export. Use redact_string() for manual control. Pattern-based today; ML-based detection on the roadmap.
Many agent tools are built for development and debugging, requiring deep inspection of prompts. Traccia is built for operations at scale.
Centralizing sensitive data creates a massive attack surface.
By design, Traccia avoids becoming a data exfiltration surface.
Prevents accidental data capture at the source.
Configurable truncation and PII redaction run inside the SDK before export.
Data encrypted in transit (TLS 1.3) and at rest (AES-256).
API keys and credentials are never written to logs.
Traccia gives you visibility into how agents behave, decide, and spend — without ever asking to see what they think or what they process.