v148 · origin trial · on-device AI

Local-First Workflow

Three real-world app flows — Code Reviewer, Customer Reply, Brainstorm — each showing how temperature and topK should be tuned for the task. Dial the parameters, run the prompt, and see how the model's tone and diversity change. All processing stays on-device.

Origin trial — requires Chrome 148+ with the Prompt API enabled. Simulation mode is active when the API is absent.
Code Reviewer — use low temperature for consistent, precise feedback

Code review needs deterministic, repeatable analysis. Low temperature + moderate topK keeps the model focused.

Deterministic (t=0.1, k=10) Balanced (t=0.5, k=40) Creative (t=1.0, k=128)
temperature 0.1
topK 10
Output will appear here…
Customer Reply — balanced temperature for professional yet natural language

Customer-facing replies need consistency but not rigidity. Mid-range temperature + topK produces professional, varied phrasing.

Stiff (t=0.1, k=10) Natural (t=0.7, k=40) Too casual (t=1.5, k=128)
temperature 0.7
topK 40
Output will appear here…
Brainstorm — high temperature for diverse, creative ideas

Ideation benefits from diversity. High temperature + high topK lets the model explore unexpected associations.

Focused (t=0.5, k=20) Balanced (t=1.0, k=60) Wild (t=1.8, k=128)
temperature 1.8
topK 128
Output will appear here…

implementation reference

Need the exact API surface, compatibility boundaries, errors, lifecycle, and source links? Read the matching gendn reference ↗