dsh-live-model-catalog

DSH's model discovery has three common gaps: builtin providers (openrouter, deepseek, etc.) return a static bundled snapshot instead of hitting the network, hand-written custom routes lack reasoning-effort levels, and some endpoints expose only model ids without any capability metadata to inherit. dsh-live-model-catalog closes all three by polling each route's own endpoint and writing the live model list along with contextWindow, maxTokens, input, and reasoningEfforts into DSH settings, keeping every route in sync with what the endpoint actually serves.

The plugin writes exclusively through DSH's official settings seam — no patches, no source modifications, no forking of pi-ai. Every write carries an expectedRevision for safe concurrency, and existing values are never overwritten. Endpoint-declared capabilities always take priority; silent fields are filled from your one-time declaration, and the plugin maintains them across sync cycles. It also takes over the "fetch available models" button so that even builtin providers not yet added to settings produce a live listing. Every outcome — written, unchanged, skipped, or failed — lands in a per-model report, so troubleshooting is never a guessing game.

Built for DSH users who rely on builtin-provider routes and want timely access to new models with full capability descriptions; developers maintaining custom routes that need reasoning-effort selectors in the composer; and teams whose endpoints (such as SenseNova) ship sparse model lists requiring manual effort declarations. If your DSH workflow keeps hitting stale model lists, missing level pickers, or half-populated model entries, this plugin was designed for exactly that.