Skip to main content

​
Fast mode (verdict-only)

Fast mode returns the conversation-level verdict without the per-message message_analysis breakdown. Because the per-message array is the largest part of the response, omitting it cuts output length and latency substantially, which makes fast mode the right choice for screening live chat in real time. Fast mode changes only how much detail comes back. The verdict itself, the risk level, the categories, and the recommended action are unchanged.
Recommended default for programmatic integrations. The standard response format — markdown-formatted rationale, emoji severity headers, a full helpline block on every positive hit — is built for a human reading it, not for a pipeline branching on it. If your code keys logic on risk_score / severity / recommended_action and doesn’t render the response for a person, default to verdict_only: true rather than reaching for it only when latency matters.

​
When to use it

  • Programmatic/pipeline integrations. Any caller keying decisions off risk_score, severity, or recommended_action rather than rendering the response for a human — this is the default you want, not just an optimization for high volume.
  • Live chat screening. Score every message as it is sent, with minimal latency, and act on the verdict.
  • High-volume pipelines. Reduce response size and processing time across large streams of messages.
  • First pass of a two-pass flow. Screen everything in fast mode, then re-run only the messages that flag through the standard endpoint to get the full per-message analysis for your moderators.
Use standard mode when a human is going to review the case and needs the per-message trajectory and evidence — the markdown, emoji, and helpline formatting are aimed at that reader.

​
How to enable it

Set options.verdict_only to true on any detection endpoint.

​
What you get back

Fast mode returns the same verdict fields as standard mode:
  • grooming_risk (or the equivalent risk level for the endpoint)
  • risk_score and confidence
  • flags (the categories present)
  • recommended_action (the routing key your alerting should branch on)
  • action_detail (human-readable guidance for a moderator, when present)
  • rationale
The only field omitted is message_analysis, the per-message risk-and-flag array. Everything you need to make a decision is still present. A common production setup is two passes:
  1. Screen every message in fast mode. Route on recommended_action.
  2. Review the small subset that flags by re-running it through the standard endpoint, so a moderator sees the full per-message trajectory and evidence.
This gives you real-time coverage on the whole stream while keeping rich detail exactly where a human needs it.

​
Frequently asked questions

Does fast mode change the verdict? No. It only omits the per-message breakdown. The risk level, categories, and recommended action are the same as standard mode. Which endpoints support fast mode? The detection endpoints accept options.verdict_only. Use it wherever you are screening rather than reviewing. Is it suitable for live game and social chat? Yes. Low latency and a compact response make it the right mode for real-time moderation of high-volume chat. Do I still get the categories and recommended action? Yes. Only the per-message array is omitted; the verdict, flags, and recommended action remain.