Deferred idea (user, 2026-09-25)
Route by what models are good at, not only by subscription usage: for example Astra for browser work, Claude for visual work, Grok for critique and insights.
Sketch, not a design
- Hard capabilities (vision, browser or computer use, web search, long context, maximum effort) come from facts Chromeria already knows (provider snapshot, harness tools) and filter eligible models.
- Soft strengths (critique, visual or front-end design, debugging, research, writing) are tags the user sets on the Prism settings page, later backed by Agent Observer evidence (acceptance per model per need), and only rank models within a role and lane.
- The planner passes a short
needs list; no model self-descriptions; a small fixed vocabulary in the glossary.
Why deferred
The user judged it too complicated for now; roles, lanes and fallbacks (#115) come first. Revisit after toolboxmd/model-router#115's streams land and #108 has produced routing evidence.
Deferred idea (user, 2026-09-25)
Route by what models are good at, not only by subscription usage: for example Astra for browser work, Claude for visual work, Grok for critique and insights.
Sketch, not a design
needslist; no model self-descriptions; a small fixed vocabulary in the glossary.Why deferred
The user judged it too complicated for now; roles, lanes and fallbacks (#115) come first. Revisit after toolboxmd/model-router#115's streams land and #108 has produced routing evidence.