Model catalog knowledge
OKF v0.2 bundle for Kai’s curated model metadata and Arena Elo scores. Agents use this folder for matching policy, provenance, and freshness when refreshing the runtime catalog.
Runtime vs knowledge
| Layer | Path | Role |
|---|---|---|
| Runtime (shipped) | composeApp/src/commonMain/kotlin/com/inspiredandroid/kai/data/ModelCatalog.kt |
Source of truth for display name, context window, release date, params, and Elo shown in the app |
| Knowledge (this bundle) | docs/knowledge/model-catalog/ |
Matching policy, attested Arena snapshot, stale dates |
| Refresh skill | .grok/skills/update-model-catalog/SKILL.md |
Fetches the text leaderboard, updates Kotlin + this bundle |
The app never reads this directory at runtime. No live Arena calls are made in production.
Scope
- In: Arena text / overall Elo, and how those numbers map onto catalog ids.
- Out:
baseEntrieseditorial metadata (display name, context window, release date, parameter count) stays in Kotlin only — it does not churn from the leaderboard the way Elo does. - Out: Free-tier badges. Those are per-service and live in
docs/knowledge/free-tier/. Never write free-ness into this catalog.
Catalogs
- Arena text Elo scores — attested leaderboard snapshot vs estimated fills
Policy
- Matching and estimate policy — exact / punctuation aliases, what must not be copied, how gaps are filled
Playbooks
- Refresh model catalog — end-to-end update procedure (also
/update-model-catalog)