Docker Advanced Configuration¶
This page covers the advanced runtime configuration pattern for custom OpenAI-compatible models in Docker deployments.
Extra Request Parameters for Custom Models¶
When a provider needs more than apiKey, baseURL, and model, use:
VITE_CUSTOM_API_PARAMS_<suffix>=json-object-string
Example:
services:
prompt-optimizer:
image: linshen/prompt-optimizer:latest
environment:
VITE_CUSTOM_API_KEY_nvidia: nvapi-xxx
VITE_CUSTOM_API_BASE_URL_nvidia: https://integrate.api.nvidia.com/v1
VITE_CUSTOM_API_MODEL_nvidia: qwen/qwen3.5-397b-a17b
VITE_CUSTOM_API_PARAMS_nvidia: '{"chat_template_kwargs":{"enable_thinking":true},"temperature":0.6,"top_p":0.95,"max_tokens":16384}'
What It Supports¶
- standard request fields such as
temperature,top_p, andmax_tokens - vendor-specific fields such as NVIDIA NIM's
chat_template_kwargs - stable runtime defaults supplied by Docker instead of manual UI input
Constraints¶
PARAMSmust be a JSON object string- reserved keys
model,messages, andstreamare ignored automatically timeoutis allowed and can be used to override request timeout- invalid JSON does not break the model configuration; only the extra params are ignored
Verification¶
- Start the container.
- Pick the custom model in the UI.
- Send a message.
- Inspect the outgoing request body in browser DevTools.