Short intro
Zenquanta becomes more interesting when assistants are product primitives instead of renamed model dropdowns.
What I was trying to do
I wanted assistant families to organize the product: Nova, Velora, Axiom, Forge, Pulse, and Prism should imply different jobs, routing policies, and expectations.
What I learned
- Assistant families create product structure.
- Prompt precheck improves UX before the expensive model call.
- Model routing needs plan awareness and admin controls.
- Usage tracking needs to separate raw backend cost from user-facing usage.
- Admin activation and plan requests are part of the product, not admin afterthoughts.
Technical notes
- Nova, Velora, Axiom, Forge, Pulse, and Prism can map to default tasks and model families.
- Prompt precheck can recommend assistant changes, identify unsupported modalities, and flag risky input.
- ModelCall and UsageRecord should be separate because one backend call can map to different user-facing accounting.
- Admin model activation should leave an audit trail.
Problems / open questions
- How visible should assistant routing be to users?
- Should plan requests be manual, automatic, or hybrid?
- Which assistant families need memory first?
Next steps
- Document assistant family responsibilities.
- Add admin views for model activation and call history.
- Connect prompt precheck results to assistant recommendations.
- Tie usage display to plan limits.