qingshanjiluo/dsh-model-caps
Fetch a provider's real model list, see which capability fields it actually exposes, edit them and save them back into llm-pi-ai.
Fills in per-model capability fields that llm-pi-ai accepts but most OpenAI-compatible gateways never return. You pick a provider, fetch its upstream model list with that provider's key, and get a table where fields sourced from the upstream payload are marked and blank cells mean the upstream did not expose the value — nothing is guessed from a static name table, which is the point for private routes. Vision support can be proven rather than assumed: the probe sends a 32x32 red and blue PNG and a model counts as image-capable only if it answers with a colour. You edit any cell and save, and the plugin writes back through settings.mutate while preserving fields it did not touch.
Install
dsh plugin --profile web add github:qingshanjiluo/dsh-model-capsAfter installing, restart the profile (or let dsh-hotswap mount it if that is installed) and open the editor at 127.0.0.1:3080/model-caps/ in the same browser session as the Web UI, because the cookie is what authenticates the page. A settings panel page also embeds the editor so you do not need the URL. The editor HTML is read from disk on every request, so a UI change takes effect on refresh without restarting DSH.
Compatibility
DSH Web with llm-pi-ai providers configured. The provider list comes from llm-pi-ai.providers and the API key is looked up from apiKeyEnv through the credential store, so no key is entered into the plugin itself.
Details
- Repo: qingshanjiluo/dsh-model-caps
- Category: Models & Providers
- Stars: 0
- Version: not published to npm; installs from the GitHub default branch
Alternatives
karoc/dsh-model-reasoning · hi-fangj/dsh-models-radar · lincong1987/dsh-model-switch