Added an agent_server module for better compartmentalization.

Dockerfile

@@ -56,6 +56,7 @@ COPY --from=base --chown=1000 /app/.env /app/.env
 COPY --chown=1000 .env.local /app/.env.local
 COPY --chown=1000 proxy.py /app/proxy.py
 COPY --chown=1000 agents /app/agents
+COPY --chown=1000 agent_server /app/agent_server
 COPY --chown=1000 --chmod=0755 entrypoint.sh /app/entrypoint.sh
 
 ENTRYPOINT ["/app/entrypoint.sh"]

agent_server/openai_schemas.py

@@ -0,0 +1,14 @@
+import typing
+
+
+class ChatMessage(typing.TypedDict, total=False):
+    role: str
+    content: typing.Any  # str or multimodal list
+
+
+class ChatCompletionRequest(typing.TypedDict, total=False):
+    model: typing.Optional[str]
+    messages: typing.List[ChatMessage]
+    temperature: typing.Optional[float]
+    stream: typing.Optional[bool]
+    max_tokens: typing.Optional[int]

proxy.py

@@ -19,6 +19,7 @@ import contextlib
 # Upstream pass-through
 import httpx
 
+from agent_server.openai_schemas import ChatMessage, ChatCompletionRequest
 from agents.code_writing_agents import (
     generate_code_writing_agent_without_tools,
     generate_code_writing_agent_with_search,
@@ -36,12 +37,7 @@ log = logging.getLogger(__name__)
 
 # Config from env vars
 UPSTREAM_BASE = os.getenv("UPSTREAM_OPENAI_BASE", "").rstrip("/")
-HF_TOKEN = (
-    os.getenv("HF_TOKEN")
-    or os.getenv("HUGGINGFACEHUB_API_TOKEN")
-    or os.getenv("API_TOKEN")
-    or ""
-)
+HF_TOKEN = os.getenv("OPENAI_API_KEY")
 AGENT_MODEL = os.getenv("AGENT_MODEL", "Qwen/Qwen3-1.7B")
 
 if not UPSTREAM_BASE:
@@ -60,20 +56,6 @@ async def healthz():
     return {"ok": True}
 
 
-# ---------- OpenAI-compatible minimal schemas ----------
-class ChatMessage(typing.TypedDict, total=False):
-    role: str
-    content: typing.Any  # str or multimodal list
-
-
-class ChatCompletionRequest(typing.TypedDict, total=False):
-    model: typing.Optional[str]
-    messages: typing.List[ChatMessage]
-    temperature: typing.Optional[float]
-    stream: typing.Optional[bool]
-    max_tokens: typing.Optional[int]
-
-
 # ---------- Helpers ----------
 def normalize_content_to_text(content: typing.Any) -> str:
     if isinstance(content, str):