Let Graph RAG demo use OpenAI instead of local models.

Dockerfile

@@ -13,7 +13,8 @@ RUN --mount=type=secret,id=LYNXSCRIBE_DEPLOY_KEY,mode=0444,required=true \
   -e lynxkite-graph-analytics \
   -e lynxkite-bio \
   -e lynxkite-lynxscribe \
-  -e lynxkite-pillow-example
+  -e lynxkite-pillow-example \
+  chromadb openai
 WORKDIR $HOME/app/examples
 ENV PORT=7860
 CMD ["uv", "run", "lynxkite"]

lynxkite-lynxscribe/src/lynxkite_lynxscribe/llm_ops.py

@@ -25,7 +25,7 @@ op = ops.op_registration(ENV)
 def chat(*args, **kwargs):
     import openai
 
-    chat_client = openai.OpenAI(base_url="http://localhost:8080/v1")
+    chat_client = openai.OpenAI()
     key = json.dumps({"method": "chat", "args": args, "kwargs": kwargs})
     if key not in LLM_CACHE:
         completion = chat_client.chat.completions.create(*args, **kwargs)
@@ -36,7 +36,7 @@ def chat(*args, **kwargs):
 def embedding(*args, **kwargs):
     import openai
 
-    embedding_client = openai.OpenAI(base_url="http://localhost:7997/")
+    embedding_client = openai.OpenAI()
     key = json.dumps({"method": "embedding", "args": args, "kwargs": kwargs})
     if key not in LLM_CACHE:
         res = embedding_client.embeddings.create(*args, **kwargs)