Add examples and documentation for using custom model providers

docs/models.md

@@ -53,21 +53,14 @@ async def main():
 
 ## Using other LLM providers
 
-Many providers also support the OpenAI API format, which means you can pass a `base_url` to the existing OpenAI model implementations and use them easily. `ModelSettings` is used to configure tuning parameters (e.g., temperature, top_p) for the model you select.
+You can use other LLM providers in 3 ways (examples [here](https://github.com/openai/openai-agents-python/tree/main/examples/model_providers/)):
 
-```python
+1. [`set_default_openai_client`][agents.set_default_openai_client] is useful in cases where you want to globally use an instance of `AsyncOpenAI` as the LLM client. This is for cases where the LLM provider has an OpenAI compatible API endpoint, and you can set the `base_url` and `api_key`. See a configurable example in [examples/model_providers/custom_example_global.py](https://github.com/openai/openai-agents-python/tree/main/examples/model_providers/custom_example_global.py).
-external_client = AsyncOpenAI(
+2. [`ModelProvider`][agents.models.interface.ModelProvider] is at the `Runner.run` level. This lets you say "use a custom model provider for all agents in this run". See a configurable example in [examples/model_providers/custom_example_provider.py](https://github.com/openai/openai-agents-python/tree/main/examples/model_providers/custom_example_provider.py).
-    api_key="EXTERNAL_API_KEY",
+3. [`Agent.model`][agents.agent.Agent.model] lets you specify the model on a specific Agent instance. This enables you to mix and match different providers for different agents. See a configurable example in [examples/model_providers/custom_example_agent.py](https://github.com/openai/openai-agents-python/tree/main/examples/model_providers/custom_example_agent.py).
-    base_url="https://api.external.com/v1/",
-)
 
-spanish_agent = Agent(
+In cases where you do not have an API key from `platform.openai.com`, we recommend disabling tracing via `set_tracing_disabled()`, or setting up a [different tracing processor](tracing.md).
-    name="Spanish agent",
+
-    instructions="You only speak Spanish.",
+!!! note
-    model=OpenAIChatCompletionsModel(
+
-        model="EXTERNAL_MODEL_NAME",
+    In these examples, we use the Chat Completions API/model, because most LLM providers don't yet support the Responses API. If your LLM provider does support it, we recommend using Responses.
-        openai_client=external_client,
-    ),
-    model_settings=ModelSettings(temperature=0.5),
-)
-```

examples/model_providers/README.md

@@ -0,0 +1,19 @@
+# Custom LLM providers
+
+The examples in this directory demonstrate how you might use a non-OpenAI LLM provider. To run them, first set a base URL, API key and model.
+
+```bash
+export EXAMPLE_BASE_URL="..."
+export EXAMPLE_API_KEY="..."
+export EXAMPLE_MODEL_NAME"..."
+```
+
+Then run the examples, e.g.:
+
+```
+python examples/model_providers/custom_example_provider.py
+
+Loops within themselves,
+Function calls its own being,
+Depth without ending.
+```

examples/model_providers/custom_example_agent.py

@@ -0,0 +1,51 @@
+import asyncio
+import os
+
+from openai import AsyncOpenAI
+
+from agents import Agent, OpenAIChatCompletionsModel, Runner, set_tracing_disabled
+
+BASE_URL = os.getenv("EXAMPLE_BASE_URL") or ""
+API_KEY = os.getenv("EXAMPLE_API_KEY") or ""
+MODEL_NAME = os.getenv("EXAMPLE_MODEL_NAME") or ""
+
+if not BASE_URL or not API_KEY or not MODEL_NAME:
+    raise ValueError(
+        "Please set EXAMPLE_BASE_URL, EXAMPLE_API_KEY, EXAMPLE_MODEL_NAME via env var or code."
+    )
+
+"""This example uses a custom provider for a specific agent. Steps:
+1. Create a custom OpenAI client.
+2. Create a `Model` that uses the custom client.
+3. Set the `model` on the Agent.
+
+Note that in this example, we disable tracing under the assumption that you don't have an API key
+from platform.openai.com. If you do have one, you can either set the `OPENAI_API_KEY` env var
+or call set_tracing_export_api_key() to set a tracing specific key.
+"""
+client = AsyncOpenAI(base_url=BASE_URL, api_key=API_KEY)
+set_tracing_disabled(disabled=True)
+
+# An alternate approach that would also work:
+# PROVIDER = OpenAIProvider(openai_client=client)
+# agent = Agent(..., model="some-custom-model")
+# Runner.run(agent, ..., run_config=RunConfig(model_provider=PROVIDER))
+
+
+async def main():
+    # This agent will use the custom LLM provider
+    agent = Agent(
+        name="Assistant",
+        instructions="You only respond in haikus.",
+        model=OpenAIChatCompletionsModel(model=MODEL_NAME, openai_client=client),
+    )
+
+    result = await Runner.run(
+        agent,
+        "Tell me about recursion in programming.",
+    )
+    print(result.final_output)
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

examples/model_providers/custom_example_global.py

@@ -0,0 +1,55 @@
+import asyncio
+import os
+
+from openai import AsyncOpenAI
+
+from agents import (
+    Agent,
+    Runner,
+    set_default_openai_api,
+    set_default_openai_client,
+    set_tracing_disabled,
+)
+
+BASE_URL = os.getenv("EXAMPLE_BASE_URL") or ""
+API_KEY = os.getenv("EXAMPLE_API_KEY") or ""
+MODEL_NAME = os.getenv("EXAMPLE_MODEL_NAME") or ""
+
+if not BASE_URL or not API_KEY or not MODEL_NAME:
+    raise ValueError(
+        "Please set EXAMPLE_BASE_URL, EXAMPLE_API_KEY, EXAMPLE_MODEL_NAME via env var or code."
+    )
+
+
+"""This example uses a custom provider for all requests by default. We do three things:
+1. Create a custom client.
+2. Set it as the default OpenAI client, and don't use it for tracing.
+3. Set the default API as Chat Completions, as most LLM providers don't yet support Responses API.
+
+Note that in this example, we disable tracing under the assumption that you don't have an API key
+from platform.openai.com. If you do have one, you can either set the `OPENAI_API_KEY` env var
+or call set_tracing_export_api_key() to set a tracing specific key.
+"""
+
+client = AsyncOpenAI(
+    base_url=BASE_URL,
+    api_key=API_KEY,
+)
+set_default_openai_client(client=client, use_for_tracing=False)
+set_default_openai_api("chat_completions")
+set_tracing_disabled(disabled=True)
+
+
+async def main():
+    agent = Agent(
+        name="Assistant",
+        instructions="You only respond in haikus.",
+        model=MODEL_NAME,
+    )
+
+    result = await Runner.run(agent, "Tell me about recursion in programming.")
+    print(result.final_output)
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

examples/model_providers/custom_example_provider.py

@@ -0,0 +1,73 @@
+from __future__ import annotations
+
+import asyncio
+import os
+
+from openai import AsyncOpenAI
+
+from agents import (
+    Agent,
+    Model,
+    ModelProvider,
+    OpenAIChatCompletionsModel,
+    RunConfig,
+    Runner,
+    set_tracing_disabled,
+)
+
+BASE_URL = os.getenv("EXAMPLE_BASE_URL") or ""
+API_KEY = os.getenv("EXAMPLE_API_KEY") or ""
+MODEL_NAME = os.getenv("EXAMPLE_MODEL_NAME") or ""
+
+if not BASE_URL or not API_KEY or not MODEL_NAME:
+    raise ValueError(
+        "Please set EXAMPLE_BASE_URL, EXAMPLE_API_KEY, EXAMPLE_MODEL_NAME via env var or code."
+    )
+
+
+"""This example uses a custom provider for some calls to Runner.run(), and direct calls to OpenAI for
+others. Steps:
+1. Create a custom OpenAI client.
+2. Create a ModelProvider that uses the custom client.
+3. Use the ModelProvider in calls to Runner.run(), only when we want to use the custom LLM provider.
+
+Note that in this example, we disable tracing under the assumption that you don't have an API key
+from platform.openai.com. If you do have one, you can either set the `OPENAI_API_KEY` env var
+or call set_tracing_export_api_key() to set a tracing specific key.
+"""
+client = AsyncOpenAI(base_url=BASE_URL, api_key=API_KEY)
+set_tracing_disabled(disabled=True)
+
+
+class CustomModelProvider(ModelProvider):
+    def get_model(self, model_name: str | None) -> Model:
+        return OpenAIChatCompletionsModel(model=model_name or MODEL_NAME, openai_client=client)
+
+
+CUSTOM_MODEL_PROVIDER = CustomModelProvider()
+
+
+async def main():
+    agent = Agent(
+        name="Assistant",
+        instructions="You only respond in haikus.",
+    )
+
+    # This will use the custom model provider
+    result = await Runner.run(
+        agent,
+        "Tell me about recursion in programming.",
+        run_config=RunConfig(model_provider=CUSTOM_MODEL_PROVIDER),
+    )
+    print(result.final_output)
+
+    # If you uncomment this, it will use OpenAI directly, not the custom provider
+    # result = await Runner.run(
+    #     agent,
+    #     "Tell me about recursion in programming.",
+    # )
+    # print(result.final_output)
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

src/agents/__init__.py

@@ -92,13 +92,19 @@ from .tracing import (
 from .usage import Usage
 
 
-def set_default_openai_key(key: str) -> None:
+def set_default_openai_key(key: str, use_for_tracing: bool = True) -> None:
-    """Set the default OpenAI API key to use for LLM requests and tracing. This is only necessary if
+    """Set the default OpenAI API key to use for LLM requests (and optionally tracing(). This is
-    the OPENAI_API_KEY environment variable is not already set.
+    only necessary if the OPENAI_API_KEY environment variable is not already set.
 
     If provided, this key will be used instead of the OPENAI_API_KEY environment variable.
+
+    Args:
+        key: The OpenAI key to use.
+        use_for_tracing: Whether to also use this key to send traces to OpenAI. Defaults to True
+            If False, you'll either need to set the OPENAI_API_KEY environment variable or call
+            set_tracing_export_api_key() with the API key you want to use for tracing.
     """
-    _config.set_default_openai_key(key)
+    _config.set_default_openai_key(key, use_for_tracing)
 
 
 def set_default_openai_client(client: AsyncOpenAI, use_for_tracing: bool = True) -> None:

src/agents/_config.py

@@ -5,15 +5,18 @@ from .models import _openai_shared
 from .tracing import set_tracing_export_api_key
 
 
-def set_default_openai_key(key: str) -> None:
+def set_default_openai_key(key: str, use_for_tracing: bool) -> None:
-    set_tracing_export_api_key(key)
     _openai_shared.set_default_openai_key(key)
 
+    if use_for_tracing:
+        set_tracing_export_api_key(key)
+
 
 def set_default_openai_client(client: AsyncOpenAI, use_for_tracing: bool) -> None:
+    _openai_shared.set_default_openai_client(client)
+
     if use_for_tracing:
         set_tracing_export_api_key(client.api_key)
-    _openai_shared.set_default_openai_client(client)
 
 
 def set_default_openai_api(api: Literal["chat_completions", "responses"]) -> None:

src/agents/models/openai_provider.py

@@ -38,28 +38,41 @@ class OpenAIProvider(ModelProvider):
             assert api_key is None and base_url is None, (
                 "Don't provide api_key or base_url if you provide openai_client"
             )
-            self._client = openai_client
+            self._client: AsyncOpenAI | None = openai_client
         else:
-            self._client = _openai_shared.get_default_openai_client() or AsyncOpenAI(
+            self._client = None
-                api_key=api_key or _openai_shared.get_default_openai_key(),
+            self._stored_api_key = api_key
-                base_url=base_url,
+            self._stored_base_url = base_url
-                organization=organization,
+            self._stored_organization = organization
-                project=project,
+            self._stored_project = project
-                http_client=shared_http_client(),
-            )
 
-        self._is_openai_model = self._client.base_url.host.startswith("api.openai.com")
         if use_responses is not None:
             self._use_responses = use_responses
         else:
             self._use_responses = _openai_shared.get_use_responses_by_default()
 
+    # We lazy load the client in case you never actually use OpenAIProvider(). Otherwise
+    # AsyncOpenAI() raises an error if you don't have an API key set.
+    def _get_client(self) -> AsyncOpenAI:
+        if self._client is None:
+            self._client = _openai_shared.get_default_openai_client() or AsyncOpenAI(
+                api_key=self._stored_api_key or _openai_shared.get_default_openai_key(),
+                base_url=self._stored_base_url,
+                organization=self._stored_organization,
+                project=self._stored_project,
+                http_client=shared_http_client(),
+            )
+
+        return self._client
+
     def get_model(self, model_name: str | None) -> Model:
         if model_name is None:
             model_name = DEFAULT_MODEL
 
+        client = self._get_client()
+
         return (
-            OpenAIResponsesModel(model=model_name, openai_client=self._client)
+            OpenAIResponsesModel(model=model_name, openai_client=client)
             if self._use_responses
-            else OpenAIChatCompletionsModel(model=model_name, openai_client=self._client)
+            else OpenAIChatCompletionsModel(model=model_name, openai_client=client)
         )