Add Anthropic (non-streaming) Support (#193)

Author: sydney-runkle

.github/workflows/ci.yml

@@ -95,6 +95,7 @@ jobs:
           --extra openai
           --extra vertexai
           --extra groq
+          --extra anthropic
           pytest tests/test_live.py -v
           --durations=100
         env:
@@ -103,6 +104,7 @@ jobs:
           GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}
           GOOGLE_SERVICE_ACCOUNT_CONTENT: ${{ secrets.GOOGLE_SERVICE_ACCOUNT_CONTENT }}
           GROQ_API_KEY: ${{ secrets.GROQ_API_KEY }}
+          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
 
   test:
     name: test on ${{ matrix.python-version }}

README.md

@@ -32,7 +32,7 @@ PydanticAI is a Python Agent Framework designed to make it less painful to build
 ## Why use PydanticAI
 
 * Built by the team behind Pydantic (the validation layer of the OpenAI SDK, the Anthropic SDK, LangChain, LlamaIndex, AutoGPT, Transformers, CrewAI, Instructor and many more)
-* Model-agnostic — currently OpenAI, Gemini, and Groq are supported. And there is a simple interface to implement support for other models.
+* Model-agnostic — currently OpenAI, Gemini, Anthropic, and Groq are supported. And there is a simple interface to implement support for other models.
 * [Type-safe](https://ai.pydantic.dev/agents/#static-type-checking)
 * Control flow and agent composition is done with vanilla Python, allowing you to make use of the same Python development best practices you'd use in any other (non-AI) project
 * [Structured response](https://ai.pydantic.dev/results/#structured-result-validation) validation with Pydantic

docs/api/models/anthropic.md

@@ -0,0 +1,7 @@
+# `pydantic_ai.models.anthropic`
+
+## Setup
+
+For details on how to set up authentication with this model, see [model configuration for Anthropic](../../install.md#anthropic).
+
+::: pydantic_ai.models.anthropic

docs/index.md

@@ -11,7 +11,7 @@ PydanticAI is a Python Agent Framework designed to make it less painful to build
 ## Why use PydanticAI
 
 * Built by the team behind Pydantic (the validation layer of the OpenAI SDK, the Anthropic SDK, LangChain, LlamaIndex, AutoGPT, Transformers, CrewAI, Instructor and many more)
-* Model-agnostic — currently OpenAI, Gemini, and Groq are supported, Anthropic [is coming soon](https://github.com/pydantic/pydantic-ai/issues/63). And there is a simple interface to implement support for other models.
+* Model-agnostic — currently OpenAI, Gemini, Anthropic, and Groq are supported, Anthropic [is coming soon](https://github.com/pydantic/pydantic-ai/issues/63). And there is a simple interface to implement support for other models.
 * [Type-safe](agents.md#static-type-checking)
 * Control flow and agent composition is done with vanilla Python, allowing you to make use of the same Python development best practices you'd use in any other (non-AI) project
 * [Structured response](results.md#structured-result-validation) validation with Pydantic

docs/install.md

@@ -12,6 +12,7 @@ This installs the `pydantic_ai` package, core dependencies, and libraries requir
 
 * [OpenAI API](https://platform.openai.com/docs/overview)
 * [Google VertexAI API](https://cloud.google.com/vertex-ai) for Gemini models
+* [Anthropic API](https://docs.anthropic.com/en/api/getting-started)
 * [Groq API](https://console.groq.com/docs/overview)
 
 ## Use with Pydantic Logfire
@@ -60,6 +61,12 @@ If you're using just [`VertexAIModel`][pydantic_ai.models.vertexai.VertexAIModel
 pip/uv-add 'pydantic-ai-slim[vertexai]'
 ```
 
+If you're just using [`Anthropic`][pydantic_ai.models.anthropic.AnthropicModel], run:
+
+```bash
+pip/uv-add 'pydantic-ai-slim[anthropic]'
+```
+
 To use just [`GroqModel`][pydantic_ai.models.groq.GroqModel], run:
 
 ```bash
@@ -277,6 +284,53 @@ agent = Agent(model)
 
 [`VertexAiRegion`][pydantic_ai.models.vertexai.VertexAiRegion] contains a list of available regions.
 
+### Anthropic
+
+To use [Anthropic](https://docs.anthropic.com/en/home) through their API, go to [console.anthropic.com/settings/keys](https://console.anthropic.com/settings/keys) to generate an API key.
+
+[`AnthropicModelName`][pydantic_ai.models.anthropic.AnthropicModelName] contains a list of available Anthropic models.
+
+#### Environment variable
+
+Once you have the API key, you can set it as an environment variable:
+
+```bash
+export ANTHROPIC_API_KEY='your-api-key'
+```
+
+You can then use [`AnthropicModel`][pydantic_ai.models.anthropic.AnthropicModel] by name:
+
+```py title="anthropic_model_by_name.py"
+from pydantic_ai import Agent
+
+agent = Agent('claude-3-5-sonnet-latest')
+...
+```
+
+Or initialise the model directly with just the model name:
+
+```py title="anthropic_model_init.py"
+from pydantic_ai import Agent
+from pydantic_ai.models.anthropic import AnthropicModel
+
+model = AnthropicModel('claude-3-5-sonnet-latest')
+agent = Agent(model)
+...
+```
+
+#### `api_key` argument
+
+If you don't want to or can't set the environment variable, you can pass it at runtime via the [`api_key` argument][pydantic_ai.models.anthropic.AnthropicModel.__init__]:
+
+```py title="anthropic_model_api_key.py"
+from pydantic_ai import Agent
+from pydantic_ai.models.anthropic import AnthropicModel
+
+model = AnthropicModel('claude-3-5-sonnet-latest', api_key='your-api-key')
+agent = Agent(model)
+...
+```
+
 ### Groq
 
 To use [Groq](https://groq.com/) through their API, go to [console.groq.com/keys](https://console.groq.com/keys) and follow your nose until you find the place to generate an API key.

mkdocs.yml

@@ -37,6 +37,7 @@ nav:
     - api/result.md
     - api/messages.md
     - api/exceptions.md
+    - api/models/anthropic.md
     - api/models/base.md
     - api/models/openai.md
     - api/models/ollama.md

pydantic_ai_examples/pyproject.toml

@@ -34,7 +34,7 @@ classifiers = [
 ]
 requires-python = ">=3.9"
 dependencies = [
-    "pydantic-ai-slim[openai,vertexai,groq]==0.0.12",
+    "pydantic-ai-slim[openai,vertexai,groq,anthropic]==0.0.12",
     "asyncpg>=0.30.0",
     "fastapi>=0.115.4",
     "logfire[asyncpg,fastapi,sqlite3]>=2.6",

pydantic_ai_slim/pydantic_ai/_utils.py

@@ -15,6 +15,7 @@
 from typing_extensions import ParamSpec, TypeAlias, TypeGuard, is_typeddict
 
 if TYPE_CHECKING:
+    from .messages import RetryPrompt, ToolCall, ToolReturn
     from .tools import ObjectJsonSchema
 
 _P = ParamSpec('_P')
@@ -254,3 +255,9 @@ def sync_anext(iterator: Iterator[T]) -> T:
 
 def now_utc() -> datetime:
     return datetime.now(tz=timezone.utc)
+
+
+def guard_tool_call_id(t: ToolCall | ToolReturn | RetryPrompt, model_source: str) -> str:
+    """Type guard that checks a `tool_call_id` is not None both for static typing and runtime."""
+    assert t.tool_call_id is not None, f'{model_source} requires `tool_call_id` to be set: {t}'
+    return t.tool_call_id

pydantic_ai_slim/pydantic_ai/messages.py

@@ -149,7 +149,7 @@ class ArgsDict:
 
 @dataclass
 class ToolCall:
-    """Either a tool call from the agent."""
+    """A tool call from the agent."""
 
     tool_name: str
     """The name of the tool to call."""
@@ -166,8 +166,8 @@ def from_json(cls, tool_name: str, args_json: str, tool_call_id: str | None = No
         return cls(tool_name, ArgsJson(args_json), tool_call_id)
 
     @classmethod
-    def from_dict(cls, tool_name: str, args_dict: dict[str, Any]) -> ToolCall:
-        return cls(tool_name, ArgsDict(args_dict))
+    def from_dict(cls, tool_name: str, args_dict: dict[str, Any], tool_call_id: str | None = None) -> ToolCall:
+        return cls(tool_name, ArgsDict(args_dict), tool_call_id)
 
     def has_content(self) -> bool:
         if isinstance(self.args, ArgsDict):

pydantic_ai_slim/pydantic_ai/models/__init__.py

@@ -67,6 +67,9 @@
     'ollama:qwen2',
     'ollama:qwen2.5',
     'ollama:starcoder2',
+    'claude-3-5-haiku-latest',
+    'claude-3-5-sonnet-latest',
+    'claude-3-opus-latest',
     'test',
 ]
 """Known model names that can be used with the `model` parameter of [`Agent`][pydantic_ai.Agent].
@@ -275,6 +278,10 @@ def infer_model(model: Model | KnownModelName) -> Model:
         from .ollama import OllamaModel
 
         return OllamaModel(model[7:])
+    elif model.startswith('claude'):
+        from .anthropic import AnthropicModel
+
+        return AnthropicModel(model)
     else:
         raise UserError(f'Unknown model: {model}')