Skip to content

Doc-401Pilot guide

A bounded path to production evidence.

This guide defines the proposed four-week design-partner motion. Final scope and timing are agreed with each participant before access is enabled.

1.0

Scope and prerequisites

  • One named agent workflow and one technical owner.
  • A representative test set or traffic sample with sensitive values removed.
  • The ability to call a HTTPS API from the runtime boundary.
  • Agreement on fail-open or fail-closed behavior before production traffic.
2.0

Suggested sequence

  1. Week 0 · Scope

    Choose one agent workflow, baseline volume and latency, define enforcement ownership, and agree on success criteria.

  2. Week 1 · Integrate

    Create scoped API keys, place inspection before the selected model/tool boundary, and verify allow, flag, and block handling.

  3. Week 2–3 · Observe

    Review aggregate telemetry, investigate representative findings, measure latency, and tune the consuming workflow’s policy.

  4. Week 4 · Decide

    Compare results against the baseline and document a go, extend, or stop decision.

3.0

Success criteria

Each pilot sets numeric targets. The default scorecard covers integration completion, p95 inspection latency, useful finding rate, false-positive review, blocked high-risk test cases, and operator effort. No universal detection or latency outcome is promised.

4.0

Data handling

The inspection service evaluates submitted prompt, response, tool-call, or metadata content in request memory. Raw inspection content is not written to the application database. Aggregate verdict, category counts, latency, and request identifiers are stored for telemetry. Do not submit regulated or production-sensitive data until contractual and technical controls have been agreed.

5.0

Available now vs. pilot scope

Available now

Scoped API keys, synchronous runtime inspection, structured findings, aggregate dashboard telemetry, local CLI scanning, and GitHub repository scans.

Not generally available

SSO, automated webhook/Slack alerts, formal compliance reports, custom deployment, and enterprise billing are roadmap or separately scoped work.
6.0

Support and exit

Use the agreed pilot channel for operational issues and include request IDs, timestamps, and impact without raw secrets. Email hello@otonmi.com if that channel is unavailable. Either side may stop the evaluation; participants receive an outcome summary and can request deletion of pilot application records.

Apply for the pilot