The short answer

AgentPlat includes a source-only optional adapter for TypeSafe Jev. Applications can use typed assessment signals to classify or route work, while retaining their own data selection, rubric mapping, policy and approval path. The adapter is not distributed on npm and has no published live quality or cost comparison.

What Jev adds to an agent application

AgentPlat's optional TypeSafe Jev assessor connects a Jev evaluation result to a typed AgentPlat decision. This gives an application a structured signal it can use to classify a result, request another review or route a work item to a different path. The value is in making an assessment usable by application code, rather than leaving it as unstructured commentary.

This is an integration point for developers who choose Jev. AgentPlat's local rules assessor remains available, so applications do not need Jev to use AgentPlat. Start with the Jev assessor guide and the Inference Control guide to see how the adapter fits the existing assessment and provider boundaries.

Your application chooses what gets assessed

Applications differ in what counts as a useful assessment. A product may assess whether a support response follows its policy, whether a proposed change meets a team rubric or whether an artifact needs a person to review it. The application author selects the relevant input, defines its rubric, and maps the returned signal to the application's typed decision vocabulary.

That choice matters because an assessment service cannot infer every product's obligations. Send only the artifacts and context required by your use case, set an appropriate provider key and spend limit, and document the mapping your application applies. AgentPlat's integration does not choose a universal rubric or silently opt every AgentPlat adopter into Jev.

Assessment does not grant execution authority

A Jev decision can inform a workflow, but it does not itself approve an external effect. Your AgentPlat application still owns the policy that decides whether work may proceed, whether a person must review it and whether a provider can perform a protected action. Keep the assessment result as evidence for that decision, and enforce approvals at the existing execution boundary.

For example, an application might ask Jev to assess a draft and then use the typed result to choose between ordinary review and an escalated review queue. A separate human approval and execution checkpoint can still be required before sending the draft. The result should not be described as a safety guarantee or as a replacement for your application's controls.

Current availability and evidence

The AgentPlat integration is implemented in source on the optional Jev branch, but it is not currently published as an npm package. Developers can inspect the adapter source and its implementation plan. This lets adopters review the interface without implying that an npm install or a TypeSafe account is already available.

No live AgentPlat quality, latency or cost comparison has been published. The company has not provided the API key needed to run the service-backed study, and no adjudicated application dataset is available for such a comparison. Treat the integration as an optional source preview; use your own representative evaluation and human review before relying on its outputs.

Sources and further reading

Documentation reviewed . Consult the linked documentation for current implementation details.