NewOpenMuse Generative UI
A Jev adapter chooses presentation controls and scores candidate fit for a mobile assistant.
Added to Jevfast
NewHow it uses Jev
OpenMuse supplies the user's message, a task summary, context, candidate options, and allowed controls to Jev. Jev chooses whether the assistant should use comparison cards or prose and returns a Score for each candidate's fit. The adapter validates the answers and keeps the agent's existing control unless Jev's control confidence is high enough.
What Jev decides
Choose response presentation and score a candidateExample answers · not a recorded Jev response · Source ↗
Question 1 · control
YOUR APP
INSTRUCTION
Choose a response control; comparison means comparison cards and agent means prose.
STATE
Repository-owned adapter test: user message 'Something hands-on', agent summary 'compare', context 'source', one candidate option a, and allowed controls comparison or agent. This example illustrates the request shape, not captured output.
JEV · CHOICE
- comparison
- agent
Question 2 · fit 0
YOUR APP
INSTRUCTION
Score the candidate at options[0] against the user's request using the adapter's four-step fit rubric.
STATE
Repository-owned adapter test: user message 'Something hands-on', agent summary 'compare', context 'source', one candidate option a, and allowed controls comparison or agent. This example illustrates the request shape, not captured output.
JEV · SCORE
0
App workflow
- Prepare candidate context
Pass the user's own message, summary, context, candidate options, and allowed controls to the adapter.
- Ask Jev
Request a control Choice and a fit Score for each candidate position.
- Validate and apply
Reject malformed controls or scores outside the 0–3 rubric; retain the agent's control unless Jev confidence meets the adapter rule.
Why it is interesting
It combines a discrete presentation choice with per-candidate continuous fit scores, while keeping the user's own message in state and not treating candidate labels as instructions.