Add Cloudflare AI Gateway provider

docs/api/providers.md

@@ -34,6 +34,8 @@
 
 ::: pydantic_ai.providers.vercel.VercelProvider
 
+::: pydantic_ai.providers.cloudflare.CloudflareProvider
+
 ::: pydantic_ai.providers.huggingface.HuggingFaceProvider
 
 ::: pydantic_ai.providers.moonshotai.MoonshotAIProvider

docs/models/openai.md

@@ -441,6 +441,67 @@ agent = Agent(model)
 ...
 ```
 
+### Cloudflare AI Gateway
+
+To use [Cloudflare AI Gateway](https://developers.cloudflare.com/ai-gateway/), first set up a gateway in your [Cloudflare dashboard](https://dash.cloudflare.com/?to=/:account/ai/ai-gateway) and obtain your account ID and gateway ID.
+
+!!! note
+    This provider uses Cloudflare's [unified API endpoint](https://developers.cloudflare.com/ai-gateway/usage/chat-completion/) for routing requests to multiple AI providers. For the full list of supported providers, see [Cloudflare's documentation](https://developers.cloudflare.com/ai-gateway/usage/chat-completion/#supported-providers).
+
+You can set the `CLOUDFLARE_ACCOUNT_ID`, `CLOUDFLARE_GATEWAY_ID`, and optionally `CLOUDFLARE_AI_GATEWAY_AUTH` environment variables and use the `cloudflare:` model name prefix:
+
+```python test="skip - requires actual API keys"
+from pydantic_ai import Agent
+
+# Set via environment or in code:
+# CLOUDFLARE_ACCOUNT_ID='your-account-id'
+# CLOUDFLARE_GATEWAY_ID='your-gateway-id'
+# OPENAI_API_KEY='your-openai-api-key'
+
+agent = Agent('cloudflare:openai/gpt-4o')
+...
+```
+
+Or use [`CloudflareProvider`][pydantic_ai.providers.cloudflare.CloudflareProvider] directly:
+
+```python
+from pydantic_ai import Agent
+from pydantic_ai.models.openai import OpenAIChatModel
+from pydantic_ai.providers.cloudflare import CloudflareProvider
+
+model = OpenAIChatModel(
+    'openai/gpt-4o',
+    provider=CloudflareProvider(
+        account_id='your-account-id',
+        gateway_id='your-gateway-id',
+        api_key='your-openai-api-key',
+    ),
+)
+agent = Agent(model)
+...
+```
+
+For authenticated gateways with stored API keys in Cloudflare's dashboard:
+
+```python
+from pydantic_ai import Agent
+from pydantic_ai.models.openai import OpenAIChatModel
+from pydantic_ai.providers.cloudflare import CloudflareProvider
+
+model = OpenAIChatModel(
+    'anthropic/claude-3-5-sonnet',
+    provider=CloudflareProvider(
+        account_id='your-account-id',
+        gateway_id='your-gateway-id',
+        gateway_auth_token='your-gateway-token',
+    ),
+)
+agent = Agent(model)
+...
+```
+
+See [`CloudflareProvider`][pydantic_ai.providers.cloudflare.CloudflareProvider] for additional configuration options including BYOK modes and authenticated gateways.
+
 ### Grok (xAI)
 
 Go to [xAI API Console](https://console.x.ai/) and create an API key.

docs/models/overview.md

@@ -20,6 +20,7 @@ In addition, many providers are compatible with the OpenAI API, and can be used
 - [Ollama](openai.md#ollama)
 - [OpenRouter](openai.md#openrouter)
 - [Vercel AI Gateway](openai.md#vercel-ai-gateway)
+- [Cloudflare AI Gateway](openai.md#cloudflare-ai-gateway)
 - [Perplexity](openai.md#perplexity)
 - [Fireworks AI](openai.md#fireworks-ai)
 - [Together AI](openai.md#together-ai)

pydantic_ai_slim/pydantic_ai/models/__init__.py

@@ -127,6 +127,16 @@
         'cerebras:qwen-3-32b',
         'cerebras:qwen-3-coder-480b',
         'cerebras:qwen-3-235b-a22b-thinking-2507',
+        'cloudflare:anthropic/claude-3-5-sonnet',
+        'cloudflare:cohere/command-r-plus',
+        'cloudflare:deepseek/deepseek-chat',
+        'cloudflare:google/gemini-2.0-flash',
+        'cloudflare:groq/llama-3.3-70b-versatile',
+        'cloudflare:mistral/mistral-large-latest',
+        'cloudflare:openai/gpt-4o',
+        'cloudflare:perplexity/llama-3.1-sonar-small-128k-online',
+        'cloudflare:workers-ai/@cf/meta/llama-3.1-8b-instruct',
+        'cloudflare:xai/grok-2-1212',
         'cohere:c4ai-aya-expanse-32b',
         'cohere:c4ai-aya-expanse-8b',
         'cohere:command-nightly',
@@ -675,6 +685,7 @@ def infer_model(model: Model | KnownModelName | str) -> Model:  # noqa: C901
         'azure',
         'deepseek',
         'cerebras',
+        'cloudflare',
         'fireworks',
         'github',
         'grok',

pydantic_ai_slim/pydantic_ai/models/openai.py

@@ -272,6 +272,7 @@ def __init__(
             'azure',
             'deepseek',
             'cerebras',
+            'cloudflare',
             'fireworks',
             'github',
             'grok',
@@ -329,6 +330,7 @@ def __init__(
             'azure',
             'deepseek',
             'cerebras',
+            'cloudflare',
             'fireworks',
             'github',
             'grok',

pydantic_ai_slim/pydantic_ai/profiles/perplexity.py

@@ -0,0 +1,8 @@
+from __future__ import annotations as _annotations
+
+from . import ModelProfile
+
+
+def perplexity_model_profile(model_name: str) -> ModelProfile | None:
+    """Get the model profile for a Perplexity model."""
+    return None

pydantic_ai_slim/pydantic_ai/providers/__init__.py

@@ -69,6 +69,10 @@ def infer_provider_class(provider: str) -> type[Provider[Any]]:  # noqa: C901
         from .vercel import VercelProvider
 
         return VercelProvider
+    elif provider == 'cloudflare':
+        from .cloudflare import CloudflareProvider
+
+        return CloudflareProvider
     elif provider == 'azure':
         from .azure import AzureProvider
 

pydantic_ai_slim/pydantic_ai/providers/cerebras.py

@@ -23,6 +23,35 @@
     ) from _import_error
 
 
+def cerebras_provider_model_profile(model_name: str) -> ModelProfile | None:
+    """Get the model profile for a model routed through Cerebras provider.
+
+    This function handles model profiling for models that use Cerebras's API,
+    and applies Cerebras-specific settings like unsupported model parameters.
+    """
+    prefix_to_profile = {'llama': meta_model_profile, 'qwen': qwen_model_profile, 'gpt-oss': harmony_model_profile}
+
+    profile = None
+    for prefix, profile_func in prefix_to_profile.items():
+        model_name = model_name.lower()
+        if model_name.startswith(prefix):
+            profile = profile_func(model_name)
+
+    # According to https://inference-docs.cerebras.ai/resources/openai#currently-unsupported-openai-features,
+    # Cerebras doesn't support some model settings.
+    unsupported_model_settings = (
+        'frequency_penalty',
+        'logit_bias',
+        'presence_penalty',
+        'parallel_tool_calls',
+        'service_tier',
+    )
+    return OpenAIModelProfile(
+        json_schema_transformer=OpenAIJsonSchemaTransformer,
+        openai_unsupported_model_settings=unsupported_model_settings,
+    ).update(profile)
+
+
 class CerebrasProvider(Provider[AsyncOpenAI]):
     """Provider for Cerebras API."""
 
@@ -39,27 +68,7 @@ def client(self) -> AsyncOpenAI:
         return self._client
 
     def model_profile(self, model_name: str) -> ModelProfile | None:
-        prefix_to_profile = {'llama': meta_model_profile, 'qwen': qwen_model_profile, 'gpt-oss': harmony_model_profile}
-
-        profile = None
-        for prefix, profile_func in prefix_to_profile.items():
-            model_name = model_name.lower()
-            if model_name.startswith(prefix):
-                profile = profile_func(model_name)
-
-        # According to https://inference-docs.cerebras.ai/resources/openai#currently-unsupported-openai-features,
-        # Cerebras doesn't support some model settings.
-        unsupported_model_settings = (
-            'frequency_penalty',
-            'logit_bias',
-            'presence_penalty',
-            'parallel_tool_calls',
-            'service_tier',
-        )
-        return OpenAIModelProfile(
-            json_schema_transformer=OpenAIJsonSchemaTransformer,
-            openai_unsupported_model_settings=unsupported_model_settings,
-        ).update(profile)
+        return cerebras_provider_model_profile(model_name)
 
     @overload
     def __init__(self) -> None: ...