Build from requirements.txt, and more accurate calls to models, allowing agent-model to use nothink

.env.local

@@ -3,7 +3,10 @@ OPENAI_BASE_URL=http://127.0.0.1:8000/v1/
 
 # The proxy forwards here (your HF endpoint):
 UPSTREAM_OPENAI_BASE=https://ay8ts6hfrqidjvwt.us-east-1.aws.endpoints.huggingface.cloud/v1
-AGENT_MODEL="Qwen/Qwen3-1.7B"
+
+# Models used by the proxy:
+MODEL_NAME="Qwen/Qwen3-1.7B"
+AGENT_MODEL="Qwen/Qwen3-1.7B-nothink"
 TASK_MODEL="Qwen/Qwen3-1.7B-nothink"
 
 PUBLIC_APP_NAME="Agent Examples"

Dockerfile

@@ -12,11 +12,10 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --no-ins
     rm -rf /var/lib/apt/lists/*
 
 # Upgrade pip and install runtime libs used by proxy.py
+COPY requirements.txt /tmp/requirements.txt
 RUN python3 -m pip install --no-cache-dir --upgrade pip && \
-    pip3 install --no-cache-dir \
-      fastapi uvicorn httpx[http2] \
-      smolagents[toolkit] litellm \
-      "pydantic>=2,<3"
+    pip3 install --no-cache-dir -r /tmp/requirements.txt && \
+    rm /tmp/requirements.txt
 
 # MongoDB
 RUN curl -fsSL https://www.mongodb.org/static/pgp/server-7.0.asc | \

agent_server/chat_completions.py

@@ -21,7 +21,7 @@ from agents.json_tool_calling_agents import (
     generate_tool_calling_agent_with_search_and_code,
 )
 
-AGENT_MODEL = os.getenv("AGENT_MODEL", "Qwen/Qwen3-1.7B")
+MODEL_NAME = os.getenv("MODEL_NAME", "Qwen/Qwen3-1.7B")
 
 
 def normalize_model_name(raw_model: typing.Union[str, dict, None]) -> str:
@@ -37,11 +37,11 @@ def normalize_model_name(raw_model: typing.Union[str, dict, None]) -> str:
 
 
 def is_upstream_passthrough(model_name: str) -> bool:
-    return model_name == AGENT_MODEL
+    return model_name == MODEL_NAME
 
 
 def is_upstream_passthrough_nothink(model_name: str) -> bool:
-    return model_name == f"{AGENT_MODEL}-nothink"
+    return model_name == f"{MODEL_NAME}-nothink"
 
 
 def apply_nothink_to_body(
@@ -53,7 +53,7 @@ def apply_nothink_to_body(
     - Appends '/nothink' to user message content
     """
     new_body: ChatCompletionRequest = dict(body)  # shallow copy is fine
-    new_body["model"] = AGENT_MODEL
+    new_body["model"] = MODEL_NAME
 
     new_messages: typing.List[ChatMessage] = []
     for msg in messages:

agent_server/models.py

@@ -7,7 +7,7 @@ def models_payload() -> dict:
     """
     Returns the /v1/models response payload.
     """
-    AGENT_MODEL = os.getenv("AGENT_MODEL", "Qwen/Qwen3-1.7B")
+    MODEL_NAME = os.getenv("MODEL_NAME", "Qwen/Qwen3-1.7B")
     now = agent_server.helpers.now_ts()
     return {
         "object": "list",
@@ -37,13 +37,13 @@ def models_payload() -> dict:
                 "owned_by": "you",
             },
             {
-                "id": AGENT_MODEL,
+                "id": MODEL_NAME,
                 "object": "model",
                 "created": now,
                 "owned_by": "upstream",
             },
             {
-                "id": f"{AGENT_MODEL}-nothink",
+                "id": f"{MODEL_NAME}-nothink",
                 "object": "model",
                 "created": now,
                 "owned_by": "upstream",

requirements.txt

@@ -0,0 +1,6 @@
+fastapi
+uvicorn
+httpx[http2]
+smolagents[toolkit]
+litellm
+pydantic>=2,<3