Route a request, select a tool, assign a support message or review text against a written policy. Start with a finite choice and a clear path for uncertain cases.
Five workflow guides explain the decision and its limits. The examples library lets you edit the inputs before requesting a result.
Use caseGive simple requests a fast path. Give harder ones room to reason.
ExploreUse caseMatch a request to an available tool, with an explicit no-match route.
Explore tool selection
Use caseDefine support queues, review ambiguous messages, and keep actions behind your policy.
Explore support classificationUse caseDistinguish product criticism, personal attacks and quoted reports with a human review route.
Explore text moderationUse caseAssess forum text and explained links without treating a relevance label as a URL safety verdict.
Explore spam triageExamplesTry a finite decision or classify a small batch, using inputs you can change.
Browse examplesThe application defines the choices and decides how to use a result. A clearer score cannot replace missing context or permission to act.
Select from model tiers, support queues, or available tools with descriptions that distinguish their jobs.
Provide the relevant context and compare the result with a label your team can independently review.
Include no match or review when appropriate, then validate routing rules before connecting a downstream action.
An open-ended explanation or reply needs a generation step. Selecting a route does not produce that reply.
A model result cannot grant access, issue a refund, or authorize an irreversible change. Those checks belong to your system.
Start with a decision you already make manually: selecting a model or tool, assigning a support owner, applying a conduct rule, or triaging relevance and promotion. Choose one policy at a time and include a review path where context is insufficient.
No. They explain the design and load editable classification examples. The Playground returns decisions; it does not run your tools, assign real tickets, remove posts or ban accounts. Integration requires application code and permission checks.
Yes. Open an example, edit the context, question, or choices, and run it through the protected API. Results are obtained when you submit, rather than stored on these pages.
No. Review errors on representative data, choose thresholds for the consequences of each action, and provide a review path. These workflow pages do not report benchmark accuracy or guaranteed savings.
Use an editable example to understand the input and output, then read the matching implementation guide.