Developer Toolsdsh-plugin
dsh-custom-provider-reasoning
A dsh host plugin that gives custom-provider models outside the pi-ai catalog selectable reasoning effort. It writes reasoningEfforts for eligible models, fills missing maxTokens/contextWindow, and manages route-level compat for openai-completions, openai-responses, and anthropic-messages.
Stars
4
Forks
0
Open issues
0
Last push
Aug 31, 2026
Latest release
—
24h Growth
+0
7d Growth
+0
Installation
No reliable install command was detected. Check the project README for installation instructions.
Overview
A dsh host plugin that gives custom-provider models outside the pi-ai catalog selectable reasoning effort. It writes reasoningEfforts for eligible models, fills missing maxTokens/contextWindow, and manages route-level compat for openai-completions, openai-responses, and anthropic-messages.
- Injects reasoningEfforts into eligible custom-provider models when no reasoningEfforts are present; user-authored values remain authoritative.
- Fills missing maxTokens (default 384000) and contextWindow (default 1000000); existing values are never overwritten.
- Protocol-aware management of route-level compat for openai-completions, openai-responses, and anthropic-messages.
- For openai-completions routes with model id containing deepseek, can add compat.thinkingFormat: deepseek when configured.
- For declared openai-completions routes, can add compat.supportsDeveloperRole: false by default; configurable.
- Injection is idempotent: once a model is covered it is not rewritten, avoiding conflicts with the settings UI.
- Config options include enabled, scope, verify, maxTokens, contextWindow, thinkingFormat, defaultEffort, supportsDeveloperRole, and levels.
- Includes an idempotent patch script for Anthropic-style /v1/models discovery in dsh-llm-pi-ai.