Overall agent

Ask Antares

The Work-site agent that routes questions to a skill, reads the right context, calls tools and returns grounded answers or resources.

100%Public baseline11/11 passed

Evaluation levels

Focused contracts for the skills Ask Antares loads while completing a request.

01Overall agent

End-to-end behavior across routing, retrieval, tools, answers, resources, multi-turn context and latency.

100%Public baseline11/11 passed

02Internal skills

Focused contracts for the skills Ask Antares loads while completing a request.

Recent runs

No runs have been recorded for this target.

Choose an evaluation set

Sets organize cases for this target; functional dimensions help authors design them.

Dataset

Run workspace

Dataset sizeSeed

Live runs call the same production Agent visitors use. Start with a small sample.

Report

Latest run

Run the frozen cases to see scores, latency and Query IDs.

Dataset

Frozen cases

Choose a scenario and synthesize a dataset.

Add your own case

Describe a real task and the behavior you expect. The case is checked for clarity and testability before it can join a run.

Functional dimension

RoutingWrite two requests that look similar but should take different paths.GroundingAsk for a factual answer whose evidence exists in one known source.Tool useDescribe a request that cannot be completed correctly without one specific tool.Resource deliveryAsk for a specific file or destination and state exactly what should be returned.Answer qualityProvide the important facts the answer must include and anything it must avoid.Multi-turn contextWrite a short conversation where the final turn depends on one earlier detail.LatencyChoose a common request and set a realistic first-token or completion budget.ReliabilityRewrite the same intent in another language or with a different phrasing.

Language Expected internal skill User request Expected behavior Answer must include Optional Answer must not include Optional Email for the report Optional

If you add an email, a report is sent after this accepted case completes a production run.

Complete the case, then run the quality review.

Cases added in this browser

No user-authored cases yet.