自定义提供商需要可行设置上下文窗口大小、推理强度、单次最大输出长度 #1316

Activity

1. henry smith (alwaysbetter1314)

   added

   enhancementNew feature or improvement (feature-request template)New feature or improvement (feature-request template)

   on Nov 11, 2026
2. danyaberezun commented on Nov 12, 2026

   

   danyaberezun

   on Nov 12, 2026

   Collaborator

   More actions

   Status update:
   Done: each model of a custom server now has an editable context window in Settings → Providers → Custom servers (the server dialog). It's stored as a modelOverrides entry, so Find models and Save keep it (#1197, v0.4.0).
   Still open:

   • Max output and the reasoning on/off flag aren't editable in the dialog. That's a deliberate choice in the current design (they're treated as facts the server reports), and it's under discussion. Reasoning effort itself is picked per chat in the model picker for models marked as reasoning.
   • Heads-up: saving the server dialog rewrites maxTokens / reasoning / input for its models. It moves them out of modelOverrides into models[] using the dialog's values, so hand edits in models.json can be lost, e.g. after Find models, or when maxTokens isn't below the context window.
     Keeping this open for the remaining part.