Data governance for AI agents
Sensitive data belongs in context — not in your logs.
Verus detects and redacts sensitive data before supported downstream systems persist it.
How it works
Detect, trace, control, ship.
Four steps, all inline with the agent — nothing round-trips to classify a value.
Detect, in-process
Sensitive values are found where the agent runs — common PII and credentials, plus your own custom patterns.
Trace provenance
Every match is connected to where it came from and where it's headed.
Apply destination-aware controls
Most values are redacted with a typed marker and moved into a vault. Secrets can be blocked from writing at all.
Ship clean
Logs, traces, memory, and evals persist without the sensitive payload.
What Verus protects
Not just the obvious fields.
Detection covers the defaults and the values only your company would recognize.
Common PII & identity
Names, emails, phone numbers, physical addresses, SSNs and other government IDs.
Credentials & secrets
API keys, access tokens, passwords, private keys — blocked outright, not just redacted.
Custom & company-specific
Internal account IDs, case numbers, proprietary terms — your own patterns, not just the defaults.
Visibility without plaintext
See what was protected, not what it was.
3 sensitive fields detected
scanned in 43ms · policy pol_crm_001Sanitized in 37ms
ready to store · all policy checks passedBuilt for agent workflows
Runs in-process, in Python, alongside what you already use.
Detection happens inline with your agent — no round-trip to classify a value.
from verus import boundary
# wrap once, where the ticket enters the run
agent = boundary(
agent,
watch=["ticket.body", "ticket.email"],
destinations=["langfuse", "otel", "eval_store"],
)
# unchanged — agent still reads and responds with the full ticket
response = agent.run(ticket)
# everything persisted downstream is redacted or blocked in placeGet started
See what your agents are actually writing.
If you're running agents in production and you're not sure what's ending up in your logs, traces, or eval data, we want to help you find out.
Or write directly: hello@verus.dev