Add LiteLLM / OpenAI-compatible proxy docs to AI config

Document how to use OpenAI-compatible proxies like LiteLLM with Lightdash
AI Analyst by pointing OPENAI_BASE_URL at the proxy endpoint. This has
come up from multiple customers (Pylon #10882).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: torimwhaley

self-host/customize-deployment/environment-variables.mdx

@@ -253,6 +253,21 @@ To enable AI Analyst, set `AI_COPILOT_ENABLED=true` and provide an API key for `
 | `OPENAI_BASE_URL`         | Optional base URL for OpenAI compatible API                                                 |
 | `OPENAI_AVAILABLE_MODELS` | Comma-separated list of models available in the model picker (default=All supported models) |
 
+<Tip>
+**Using an OpenAI-compatible proxy (e.g. LiteLLM)**
+
+If your organization uses an OpenAI-compatible proxy like [LiteLLM](https://www.litellm.ai/), you can connect it to Lightdash by setting `AI_DEFAULT_PROVIDER=openai` and pointing `OPENAI_BASE_URL` to your proxy URL. For example:
+
+- `AI_DEFAULT_PROVIDER=openai`
+- `OPENAI_API_KEY=<your-proxy-api-key>`
+- `OPENAI_BASE_URL=<your-proxy-url>`
+- `OPENAI_MODEL_NAME=<model-name-configured-in-your-proxy>`
+
+Make sure your proxy has the model names correctly mapped. For example, if you set `OPENAI_MODEL_NAME=gpt-4o`, your proxy needs to have that model routed to whatever underlying provider you're using. The same applies for the embedding model.
+
+If you want to expose multiple models through the proxy, use `OPENAI_AVAILABLE_MODELS` with a comma-separated list of model names.
+</Tip>
+
 #### Anthropic Configuration
 
 | Variable                     | Description                                                                                 |