OpenAIProvider
Field |
Type |
Repeated |
Description |
api_key |
|||
model |
|||
generation |
|||
base_url |
Optional OpenAI-compatible endpoint override, e.g. a local vLLM / llama.cpp / Ollama server ("http://127.0.0.1:11434/v1"). "" = the real OpenAI API. Enables fully-local / airgapped reasoning with permissively-licensed models (e.g. Qwen3-VL). |
||
max_tokens |
Cap on output tokens per request. 0 = no cap (the provider’s default). Worth setting for local models: one that free-runs otherwise holds a single-request server (Ollama) until the client gives up. |