PII redaction and AI data controls
Control exactly what AI providers see: masking, provider choice, retention.
Updated Aug 3, 2026
AI features involve sending workspace content to a model provider. You control what that means.
PII redaction
With Settings → Security & data → Redact PII before AI, personal identifiers are masked before any AI provider call:
- Email addresses →
[email] - Phone numbers →
[phone] - Card-like and SSN-like numbers →
[number]
Redaction happens server-side in ProcessTwin — the provider never receives the original. Masking at ingestion (connector setting) goes further: masked data never enters storage at all.
Provider choice
Pick the AI provider for your workspace (Anthropic or OpenAI). Under our agreements, your content is not used to train either provider's models.
Prompt & trace retention
Agent runs keep full traces — every prompt and response — for debugging and audit. You choose how long: 7 to 365 days, after which traces are hard-deleted by the retention sweeper. Aggregated metrics (tokens, cost, outcomes) survive; content does not.
The audit view
Curious what actually left the building? The agent trace shows the exact redacted prompt sent, token counts and provider — per run, per step.