Overview
dsh-llm-openai-completions
OpenAI-completions-compatible adapter for custom gateways (vLLM / LM Studio / self-hosted proxies): always role:"system", enable_thinking / chat_template_kwargs / reasoning_effort driven by the llm-pi-ai model config, and a split on the receive side — no pi-ai behavior guessing.
README / EN
Package documentation
Registry summary
OpenAI-completions-compatible adapter for custom gateways (vLLM / LM Studio / self-hosted proxies): always role:"system", enable_thinking / chat_template_kwargs / reasoning_effort driven by the llm-pi-ai model config, and a split on the receive side — no pi-ai behavior guessing.
dsh.pub verifies the pinned bundle contract, runtime facts, and distribution semantics. The complete README remains in the source repository.
Read the full README on GitHubLIMITATIONS
Known limitations
- **Non-vision models do not support images**: a model that does not declare image input (via `input` containing `image`) is rejected with `UNSUPPORTED_CONTENT` (image bytes live in the attachment service and are only resolved on the vision path). - **No `thinking_budget`** (deliberate: avoid truncation surprises). - The thinking-level **selector** still comes from the model's reasoning metadata (llm-pi-ai + dsh-thinking-levels' Off/On toggle); this plugin controls the wire behavior. - Restart `dsh web` after install; the wrap re-applies on `llm/adapters-updated`.