NewReMe Jev Classifier
Run typed Jev classification as a reusable ReMe workflow component.
Added to Jevfast
NewHow it uses Jev
ReMe’s optional Jev classifier accepts nonempty text or JSON state and a named mapping of typed questions, calls AgentScope’s Jev classifier, then returns the complete typed result in the step response and shares it with later steps as `classifier_result`. The demo’s route question is Choice over support teams.
What Jev decides
Customer-support route question (illustrative)Example answers · not a recorded Jev response · Source ↗
Question 1 · route
YOUR APP
INSTRUCTION
Select the team that should handle this customer request.
STATE
The config documents a customer state of `I was charged twice.` and a route Choice among three support teams.
JEV · CHOICE
- billing
- support
- other
App workflow
- Configure classifier
Select `jev` backend and configure optional model, credential, timeout, retry count, and delay.
- Submit state and questions
Call the served `jev` job with required text/JSON state; omit questions to use demo defaults, or supply a nonempty named question map.
- Call Jev
Validate state/questions, lazily load AgentScope classifier support, construct the typed classifier client, and send state plus parsed questions.
- Share result
Serialize the full classifier response into JSON, store it in context as `classifier_result`, and return it in the step answer.
Why it is interesting
The classifier is lazy-loaded: importing/configuring ReMe does not require the optional AgentScope classifier package until a classification call. The demo also shows the output as reusable workflow state, not only a one-off display.