Add LiteLLM support in line with AsyncOpenAI and with reproducible examples

examples/model_providers/README.md

@@ -17,3 +17,29 @@ Loops within themselves,
 Function calls its own being,
 Depth without ending.
 ```
+
+
+## LiteLLM Proxy Server integration
+
+LiteLLM integration helps out switch between models easily and rapidly. This is easy to integrate via `AsyncOpenAI`:
+
+```bash
+# launch server proxy with your configuration
+litellm --config examples/model_providers/litellm_config.yaml
+
+# then use the proxy via the SDK
+python3 examples/model_providers/litellm.py
+```
+
+### Testing the proxy server
+Testing some basic models against proxy to verify it's operational:
+```bash
+# qwen2.5:14b
+curl -s http://localhost:4000/v1/chat/completions -H "Content-Type: application/json" -d '{"model": "qwen2.5:14b", "messages": [{"role": "user", "content": "Say hi"}], "max_tokens": 10}' | jq
+
+# claude-3-7
+curl -s http://localhost:4000/v1/chat/completions -H "Content-Type: application/json" -d '{"model": "claude-3-7", "messages": [{"role": "user", "content": "Say hi"}], "max_tokens": 10}' | jq
+
+# gpt-4o
+curl -s http://localhost:4000/v1/chat/completions -H "Content-Type: application/json" -d '{"model": "gpt-4o", "messages": [{"role": "user", "content": "Say hi"}], "max_tokens": 10}' | jq
+```

examples/model_providers/litellm.py

@@ -0,0 +1,36 @@
+import os
+from dotenv import load_dotenv
+from openai import AsyncOpenAI
+from agents import OpenAIChatCompletionsModel,Agent,Runner
+from agents.model_settings import ModelSettings
+from agents import set_default_openai_client, set_tracing_disabled
+
+# Load environment variables from .env file
+load_dotenv()
+
+external_client = AsyncOpenAI(
+    base_url = os.getenv('LITELLM_BASE_URL', 'http://localhost:4000'),
+    api_key=os.getenv('LITELLM_API_KEY', 'key'))
+
+set_default_openai_client(external_client)
+set_tracing_disabled(True)
+
+llm_model=os.getenv('LLM_MODEL', 'gpt-4o')
+# llm_model=os.getenv('LLM_MODEL', 'claude-3-7')
+# llm_model=os.getenv('LLM_MODEL', 'qwen2.5:14b')
+
+# For Qwen models, we need to skip system instructions as they're not supported
+instructions = None if "qwen" in llm_model.lower() else "You are a helpful assistant"
+
+agent = Agent(
+    name="Assistant", 
+    instructions=instructions,
+    model=OpenAIChatCompletionsModel(
+        model=llm_model,
+        openai_client=external_client,
+    )
+)
+
+
+result = Runner.run_sync(agent, "Write a haiku about recursion in programming.")
+print(result.final_output)

examples/model_providers/litellm_config.yaml

@@ -0,0 +1,23 @@
+model_list:
+  - model_name: gpt-4o
+    litellm_params:
+      model: gpt-4o
+      api_key: "os.environ/OPENAI_API_KEY"
+      api_base: https://api.openai.com/v1
+  - model_name: claude-3-7
+    litellm_params: 
+      model: claude-3-7-sonnet-20250219
+      api_key: "os.environ/ANTHROPIC_API_KEY" # does os.getenv("ANTHROPIC_API_KEY")
+      base_url: 'https://api.anthropic.com'
+  - model_name: qwen2.5:14b
+    litellm_params:
+      model: ollama/qwen2.5:14b
+      api_base: http://localhost:8000
+      api_key: "os.environ/OLLAMA"
+
+litellm_settings: # module level litellm settings - https://github.com/BerriAI/litellm/blob/main/litellm/__init__.py
+  drop_params: True
+  # set_verbose: True
+
+general_settings: 
+  port: 4000

pyproject.toml

@@ -7,7 +7,7 @@ requires-python = ">=3.9"
 license = "MIT"
 authors = [{ name = "OpenAI", email = "[email protected]" }]
 dependencies = [
-    "openai>=1.66.5",
+    "openai>=1.68.2",
     "pydantic>=2.10, <3",
     "griffe>=1.5.6, <2",
     "typing-extensions>=4.12.2, <5",