pydantic-ai 0.2.10__tar.gz → 0.2.12__tar.gz

{pydantic_ai-0.2.10 → pydantic_ai-0.2.12}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pydantic-ai
-Version: 0.2.10
+Version: 0.2.12
 Summary: Agent Framework / shim to use Pydantic with LLMs
 Project-URL: Homepage, https://ai.pydantic.dev
 Project-URL: Source, https://github.com/pydantic/pydantic-ai
@@ -28,9 +28,9 @@ Classifier: Topic :: Internet
 Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
 Classifier: Topic :: Software Development :: Libraries :: Python Modules
 Requires-Python: >=3.9
-Requires-Dist: pydantic-ai-slim[a2a,anthropic,bedrock,cli,cohere,evals,google,groq,mcp,mistral,openai,vertexai]==0.2.10
+Requires-Dist: pydantic-ai-slim[a2a,anthropic,bedrock,cli,cohere,evals,google,groq,mcp,mistral,openai,vertexai]==0.2.12
 Provides-Extra: examples
-Requires-Dist: pydantic-ai-examples==0.2.10; extra == 'examples'
+Requires-Dist: pydantic-ai-examples==0.2.12; extra == 'examples'
 Provides-Extra: logfire
 Requires-Dist: logfire>=3.11.0; extra == 'logfire'
 Description-Content-Type: text/markdown

pydantic_ai-0.2.12/tests/models/cassettes/test_anthropic/test_anthropic_model_empty_message_on_history.yaml

@@ -0,0 +1,73 @@
+interactions:
+- request:
+    headers:
+      accept:
+      - application/json
+      accept-encoding:
+      - gzip, deflate
+      connection:
+      - keep-alive
+      content-length:
+      - '281'
+      content-type:
+      - application/json
+      host:
+      - api.anthropic.com
+    method: POST
+    parsed_body:
+      max_tokens: 1024
+      messages:
+      - content:
+        - text: Hello, how can I help you?
+          type: text
+        role: assistant
+      - content:
+        - text: I need a potato!
+          type: text
+        role: user
+      model: claude-3-5-sonnet-latest
+      stream: false
+      system: |+
+        You are a helpful assistant.
+
+    uri: https://api.anthropic.com/v1/messages?beta=true
+  response:
+    headers:
+      connection:
+      - keep-alive
+      content-length:
+      - '671'
+      content-type:
+      - application/json
+      strict-transport-security:
+      - max-age=31536000; includeSubDomains; preload
+      transfer-encoding:
+      - chunked
+    parsed_body:
+      content:
+      - text: |-
+          I can't physically give you a potato since I'm a computer program. However, I can:
+
+          1. Help you find recipes that use potatoes
+          2. Give you tips on how to select, store, or cook potatoes
+          3. Share information about different potato varieties
+          4. Provide guidance on growing potatoes
+
+          What specifically would you like to know about potatoes?
+        type: text
+      id: msg_01UjnDmX3B57Drosu49sMteT
+      model: claude-3-5-sonnet-20241022
+      role: assistant
+      stop_reason: end_turn
+      stop_sequence: null
+      type: message
+      usage:
+        cache_creation_input_tokens: 0
+        cache_read_input_tokens: 0
+        input_tokens: 41
+        output_tokens: 82
+        service_tier: standard
+    status:
+      code: 200
+      message: OK
+version: 1

{pydantic_ai-0.2.10 → pydantic_ai-0.2.12}/tests/models/test_anthropic.py

@@ -1035,3 +1035,31 @@ def anth_msg(usage: BetaUsage) -> BetaMessage:
 )
 def test_usage(message_callback: Callable[[], BetaMessage | BetaRawMessageStreamEvent], usage: Usage):
     assert _map_usage(message_callback()) == usage
+
+
+@pytest.mark.vcr()
+async def test_anthropic_model_empty_message_on_history(allow_model_requests: None, anthropic_api_key: str):
+    """The Anthropic API will error if you send an empty message on the history.
+
+    Check <https://github.com/pydantic/pydantic-ai/pull/1027> for more details.
+    """
+    m = AnthropicModel('claude-3-5-sonnet-latest', provider=AnthropicProvider(api_key=anthropic_api_key))
+    agent = Agent(m, instructions='You are a helpful assistant.')
+
+    result = await agent.run(
+        'I need a potato!',
+        message_history=[
+            ModelRequest(parts=[], instructions='You are a helpful assistant.', kind='request'),
+            ModelResponse(parts=[TextPart(content='Hello, how can I help you?')], kind='response'),
+        ],
+    )
+    assert result.output == snapshot("""\
+I can't physically give you a potato since I'm a computer program. However, I can:
+
+1. Help you find recipes that use potatoes
+2. Give you tips on how to select, store, or cook potatoes
+3. Share information about different potato varieties
+4. Provide guidance on growing potatoes
+
+What specifically would you like to know about potatoes?\
+""")

{pydantic_ai-0.2.10 → pydantic_ai-0.2.12}/tests/models/test_bedrock.py

@@ -31,6 +31,8 @@ from pydantic_ai.messages import (
     UserPromptPart,
     VideoUrl,
 )
+from pydantic_ai.models import ModelRequestParameters
+from pydantic_ai.tools import ToolDefinition
 from pydantic_ai.usage import Usage
 
 from ..conftest import IsDatetime, try_import
@@ -631,3 +633,96 @@ async def test_bedrock_group_consecutive_tool_return_parts(bedrock_provider: Bed
             },
         ]
     )
+
+
+async def test_bedrock_mistral_tool_result_format(bedrock_provider: BedrockProvider):
+    now = datetime.datetime.now()
+    req = [
+        ModelRequest(
+            parts=[
+                ToolReturnPart(tool_name='tool1', content={'foo': 'bar'}, tool_call_id='id1', timestamp=now),
+            ]
+        ),
+    ]
+
+    # Models other than Mistral support toolResult.content with text, not json
+    model = BedrockConverseModel('us.amazon.nova-micro-v1:0', provider=bedrock_provider)
+    # Call the mapping function directly
+    _, bedrock_messages = await model._map_messages(req)  # type: ignore[reportPrivateUsage]
+
+    assert bedrock_messages == snapshot(
+        [
+            {
+                'role': 'user',
+                'content': [
+                    {'toolResult': {'toolUseId': 'id1', 'content': [{'text': '{"foo":"bar"}'}], 'status': 'success'}},
+                ],
+            },
+        ]
+    )
+
+    # Mistral requires toolResult.content to hold json, not text
+    model = BedrockConverseModel('mistral.mistral-7b-instruct-v0:2', provider=bedrock_provider)
+    # Call the mapping function directly
+    _, bedrock_messages = await model._map_messages(req)  # type: ignore[reportPrivateUsage]
+
+    assert bedrock_messages == snapshot(
+        [
+            {
+                'role': 'user',
+                'content': [
+                    {'toolResult': {'toolUseId': 'id1', 'content': [{'json': {'foo': 'bar'}}], 'status': 'success'}},
+                ],
+            },
+        ]
+    )
+
+
+async def test_bedrock_anthropic_no_tool_choice(bedrock_provider: BedrockProvider):
+    my_tool = ToolDefinition(
+        'my_tool',
+        'This is my tool',
+        {'type': 'object', 'title': 'Result', 'properties': {'spam': {'type': 'number'}}},
+    )
+    mrp = ModelRequestParameters(function_tools=[my_tool], allow_text_output=False, output_tools=[])
+
+    # Models other than Anthropic support tool_choice
+    model = BedrockConverseModel('us.amazon.nova-micro-v1:0', provider=bedrock_provider)
+    tool_config = model._map_tool_config(mrp)  # type: ignore[reportPrivateUsage]
+
+    assert tool_config == snapshot(
+        {
+            'tools': [
+                {
+                    'toolSpec': {
+                        'name': 'my_tool',
+                        'description': 'This is my tool',
+                        'inputSchema': {
+                            'json': {'type': 'object', 'title': 'Result', 'properties': {'spam': {'type': 'number'}}}
+                        },
+                    }
+                }
+            ],
+            'toolChoice': {'any': {}},
+        }
+    )
+
+    # Anthropic models don't support tool_choice
+    model = BedrockConverseModel('us.anthropic.claude-3-7-sonnet-20250219-v1:0', provider=bedrock_provider)
+    tool_config = model._map_tool_config(mrp)  # type: ignore[reportPrivateUsage]
+
+    assert tool_config == snapshot(
+        {
+            'tools': [
+                {
+                    'toolSpec': {
+                        'name': 'my_tool',
+                        'description': 'This is my tool',
+                        'inputSchema': {
+                            'json': {'type': 'object', 'title': 'Result', 'properties': {'spam': {'type': 'number'}}}
+                        },
+                    }
+                }
+            ]
+        }
+    )

{pydantic_ai-0.2.10 → pydantic_ai-0.2.12}/tests/models/test_gemini.py

@@ -540,6 +540,7 @@ async def test_text_success(get_gemini_client: GetGeminiClient):
                 usage=Usage(requests=1, request_tokens=1, response_tokens=2, total_tokens=3, details={}),
                 model_name='gemini-1.5-flash-123',
                 timestamp=IsNow(tz=timezone.utc),
+                vendor_details={'finish_reason': 'STOP'},
             ),
         ]
     )
@@ -555,6 +556,7 @@ async def test_text_success(get_gemini_client: GetGeminiClient):
                 usage=Usage(requests=1, request_tokens=1, response_tokens=2, total_tokens=3, details={}),
                 model_name='gemini-1.5-flash-123',
                 timestamp=IsNow(tz=timezone.utc),
+                vendor_details={'finish_reason': 'STOP'},
             ),
             ModelRequest(parts=[UserPromptPart(content='Hello', timestamp=IsNow(tz=timezone.utc))]),
             ModelResponse(
@@ -562,6 +564,7 @@ async def test_text_success(get_gemini_client: GetGeminiClient):
                 usage=Usage(requests=1, request_tokens=1, response_tokens=2, total_tokens=3, details={}),
                 model_name='gemini-1.5-flash-123',
                 timestamp=IsNow(tz=timezone.utc),
+                vendor_details={'finish_reason': 'STOP'},
             ),
         ]
     )
@@ -585,6 +588,7 @@ async def test_request_structured_response(get_gemini_client: GetGeminiClient):
                 usage=Usage(requests=1, request_tokens=1, response_tokens=2, total_tokens=3, details={}),
                 model_name='gemini-1.5-flash-123',
                 timestamp=IsNow(tz=timezone.utc),
+                vendor_details={'finish_reason': 'STOP'},
             ),
             ModelRequest(
                 parts=[
@@ -647,6 +651,7 @@ async def test_request_tool_call(get_gemini_client: GetGeminiClient):
                 usage=Usage(requests=1, request_tokens=1, response_tokens=2, total_tokens=3, details={}),
                 model_name='gemini-1.5-flash-123',
                 timestamp=IsNow(tz=timezone.utc),
+                vendor_details={'finish_reason': 'STOP'},
             ),
             ModelRequest(
                 parts=[
@@ -666,6 +671,7 @@ async def test_request_tool_call(get_gemini_client: GetGeminiClient):
                 usage=Usage(requests=1, request_tokens=1, response_tokens=2, total_tokens=3, details={}),
                 model_name='gemini-1.5-flash-123',
                 timestamp=IsNow(tz=timezone.utc),
+                vendor_details={'finish_reason': 'STOP'},
             ),
             ModelRequest(
                 parts=[
@@ -688,6 +694,7 @@ async def test_request_tool_call(get_gemini_client: GetGeminiClient):
                 usage=Usage(requests=1, request_tokens=1, response_tokens=2, total_tokens=3, details={}),
                 model_name='gemini-1.5-flash-123',
                 timestamp=IsNow(tz=timezone.utc),
+                vendor_details={'finish_reason': 'STOP'},
             ),
         ]
     )
@@ -732,12 +739,12 @@ async def test_stream_text(get_gemini_client: GetGeminiClient):
                 'Hello world',
             ]
         )
-    assert result.usage() == snapshot(Usage(requests=1, request_tokens=2, response_tokens=4, total_tokens=6))
+    assert result.usage() == snapshot(Usage(requests=1, request_tokens=1, response_tokens=2, total_tokens=3))
 
     async with agent.run_stream('Hello') as result:
         chunks = [chunk async for chunk in result.stream_text(delta=True, debounce_by=None)]
         assert chunks == snapshot(['Hello ', 'world'])
-    assert result.usage() == snapshot(Usage(requests=1, request_tokens=2, response_tokens=4, total_tokens=6))
+    assert result.usage() == snapshot(Usage(requests=1, request_tokens=1, response_tokens=2, total_tokens=3))
 
 
 async def test_stream_invalid_unicode_text(get_gemini_client: GetGeminiClient):
@@ -769,7 +776,7 @@ async def test_stream_invalid_unicode_text(get_gemini_client: GetGeminiClient):
     async with agent.run_stream('Hello') as result:
         chunks = [chunk async for chunk in result.stream(debounce_by=None)]
         assert chunks == snapshot(['abc', 'abc€def', 'abc€def'])
-    assert result.usage() == snapshot(Usage(requests=1, request_tokens=2, response_tokens=4, total_tokens=6))
+    assert result.usage() == snapshot(Usage(requests=1, request_tokens=1, response_tokens=2, total_tokens=3))
 
 
 async def test_stream_text_no_data(get_gemini_client: GetGeminiClient):
@@ -840,7 +847,7 @@ async def test_stream_structured_tool_calls(get_gemini_client: GetGeminiClient):
     async with agent.run_stream('Hello') as result:
         response = await result.get_output()
         assert response == snapshot((1, 2))
-    assert result.usage() == snapshot(Usage(requests=2, request_tokens=3, response_tokens=6, total_tokens=9))
+    assert result.usage() == snapshot(Usage(requests=2, request_tokens=2, response_tokens=4, total_tokens=6))
     assert result.all_messages() == snapshot(
         [
             ModelRequest(parts=[UserPromptPart(content='Hello', timestamp=IsNow(tz=timezone.utc))]),
@@ -849,7 +856,7 @@ async def test_stream_structured_tool_calls(get_gemini_client: GetGeminiClient):
                     ToolCallPart(tool_name='foo', args={'x': 'a'}, tool_call_id=IsStr()),
                     ToolCallPart(tool_name='bar', args={'y': 'b'}, tool_call_id=IsStr()),
                 ],
-                usage=Usage(request_tokens=2, response_tokens=4, total_tokens=6),
+                usage=Usage(request_tokens=1, response_tokens=2, total_tokens=3, details={}),
                 model_name='gemini-1.5-flash',
                 timestamp=IsNow(tz=timezone.utc),
             ),
@@ -865,7 +872,7 @@ async def test_stream_structured_tool_calls(get_gemini_client: GetGeminiClient):
             ),
             ModelResponse(
                 parts=[ToolCallPart(tool_name='final_result', args={'response': [1, 2]}, tool_call_id=IsStr())],
-                usage=Usage(request_tokens=1, response_tokens=2, total_tokens=3),
+                usage=Usage(request_tokens=1, response_tokens=2, total_tokens=3, details={}),
                 model_name='gemini-1.5-flash',
                 timestamp=IsNow(tz=timezone.utc),
             ),
@@ -1096,9 +1103,16 @@ I need to use the `get_image` tool to see the image first.
                     ),
                     ToolCallPart(tool_name='get_image', args={}, tool_call_id=IsStr()),
                 ],
-                usage=Usage(requests=1, request_tokens=38, response_tokens=28, total_tokens=427, details={}),
+                usage=Usage(
+                    requests=1,
+                    request_tokens=38,
+                    response_tokens=28,
+                    total_tokens=427,
+                    details={'thoughts_tokens': 361, 'text_prompt_tokens': 38},
+                ),
                 model_name='gemini-2.5-pro-preview-03-25',
                 timestamp=IsDatetime(),
+                vendor_details={'finish_reason': 'STOP'},
             ),
             ModelRequest(
                 parts=[
@@ -1119,9 +1133,16 @@ I need to use the `get_image` tool to see the image first.
             ),
             ModelResponse(
                 parts=[TextPart(content='The image shows a kiwi fruit, sliced in half.')],
-                usage=Usage(requests=1, request_tokens=360, response_tokens=11, total_tokens=572, details={}),
+                usage=Usage(
+                    requests=1,
+                    request_tokens=360,
+                    response_tokens=11,
+                    total_tokens=572,
+                    details={'thoughts_tokens': 201, 'text_prompt_tokens': 102, 'image_prompt_tokens': 258},
+                ),
                 model_name='gemini-2.5-pro-preview-03-25',
                 timestamp=IsDatetime(),
+                vendor_details={'finish_reason': 'STOP'},
             ),
         ]
     )
@@ -1241,9 +1262,16 @@ async def test_gemini_model_instructions(allow_model_requests: None, gemini_api_
             ),
             ModelResponse(
                 parts=[TextPart(content='The capital of France is Paris.\n')],
-                usage=Usage(requests=1, request_tokens=13, response_tokens=8, total_tokens=21, details={}),
+                usage=Usage(
+                    requests=1,
+                    request_tokens=13,
+                    response_tokens=8,
+                    total_tokens=21,
+                    details={'text_prompt_tokens': 13, 'text_candidates_tokens': 8},
+                ),
                 model_name='gemini-1.5-flash',
                 timestamp=IsDatetime(),
+                vendor_details={'finish_reason': 'STOP'},
             ),
         ]
     )
@@ -1284,3 +1312,18 @@ async def test_gemini_additional_properties_is_true(allow_model_requests: None,
         assert result.output == snapshot(
             'I need a location dictionary to use the `get_temperature` function.  I cannot provide the temperature in Tokyo without more information.\n'
         )
+
+
+async def test_gemini_no_finish_reason(get_gemini_client: GetGeminiClient):
+    response = gemini_response(
+        _content_model_response(ModelResponse(parts=[TextPart('Hello world')])), finish_reason=None
+    )
+    gemini_client = get_gemini_client(response)
+    m = GeminiModel('gemini-1.5-flash', provider=GoogleGLAProvider(http_client=gemini_client))
+    agent = Agent(m)
+
+    result = await agent.run('Hello World')
+
+    for message in result.all_messages():
+        if isinstance(message, ModelResponse):
+            assert message.vendor_details is None

{pydantic_ai-0.2.10 → pydantic_ai-0.2.12}/tests/models/test_google.py

@@ -65,7 +65,15 @@ async def test_google_model(allow_model_requests: None, google_provider: GoogleP
 
     result = await agent.run('Hello!')
     assert result.output == snapshot('Hello there! How can I help you today?\n')
-    assert result.usage() == snapshot(Usage(requests=1, request_tokens=7, response_tokens=11, total_tokens=18))
+    assert result.usage() == snapshot(
+        Usage(
+            requests=1,
+            request_tokens=7,
+            response_tokens=11,
+            total_tokens=18,
+            details={'text_prompt_tokens': 7, 'text_candidates_tokens': 11},
+        )
+    )
     assert result.all_messages() == snapshot(
         [
             ModelRequest(
@@ -82,9 +90,16 @@ async def test_google_model(allow_model_requests: None, google_provider: GoogleP
             ),
             ModelResponse(
                 parts=[TextPart(content='Hello there! How can I help you today?\n')],
-                usage=Usage(requests=1, request_tokens=7, response_tokens=11, total_tokens=18, details={}),
+                usage=Usage(
+                    requests=1,
+                    request_tokens=7,
+                    response_tokens=11,
+                    total_tokens=18,
+                    details={'text_prompt_tokens': 7, 'text_candidates_tokens': 11},
+                ),
                 model_name='gemini-1.5-flash',
                 timestamp=IsDatetime(),
+                vendor_details={'finish_reason': 'STOP'},
             ),
         ]
     )
@@ -114,7 +129,15 @@ async def test_google_model_structured_response(allow_model_requests: None, goog
 
     result = await agent.run('What was the temperature in London 1st January 2022?', output_type=Response)
     assert result.output == snapshot({'temperature': '30°C', 'date': datetime.date(2022, 1, 1), 'city': 'London'})
-    assert result.usage() == snapshot(Usage(requests=2, request_tokens=224, response_tokens=35, total_tokens=259))
+    assert result.usage() == snapshot(
+        Usage(
+            requests=2,
+            request_tokens=224,
+            response_tokens=35,
+            total_tokens=259,
+            details={'text_prompt_tokens': 224, 'text_candidates_tokens': 35},
+        )
+    )
     assert result.all_messages() == snapshot(
         [
             ModelRequest(
@@ -135,9 +158,16 @@ async def test_google_model_structured_response(allow_model_requests: None, goog
                         tool_name='temperature', args={'date': '2022-01-01', 'city': 'London'}, tool_call_id=IsStr()
                     )
                 ],
-                usage=Usage(requests=1, request_tokens=101, response_tokens=14, total_tokens=115, details={}),
+                usage=Usage(
+                    requests=1,
+                    request_tokens=101,
+                    response_tokens=14,
+                    total_tokens=115,
+                    details={'text_prompt_tokens': 101, 'text_candidates_tokens': 14},
+                ),
                 model_name='gemini-1.5-flash',
                 timestamp=IsDatetime(),
+                vendor_details={'finish_reason': 'STOP'},
             ),
             ModelRequest(
                 parts=[
@@ -154,9 +184,16 @@ async def test_google_model_structured_response(allow_model_requests: None, goog
                         tool_call_id=IsStr(),
                     )
                 ],
-                usage=Usage(requests=1, request_tokens=123, response_tokens=21, total_tokens=144, details={}),
+                usage=Usage(
+                    requests=1,
+                    request_tokens=123,
+                    response_tokens=21,
+                    total_tokens=144,
+                    details={'text_prompt_tokens': 123, 'text_candidates_tokens': 21},
+                ),
                 model_name='gemini-1.5-flash',
                 timestamp=IsDatetime(),
+                vendor_details={'finish_reason': 'STOP'},
             ),
             ModelRequest(
                 parts=[
@@ -211,10 +248,11 @@ async def test_google_model_retry(allow_model_requests: None, google_provider: G
                     request_tokens=57,
                     response_tokens=15,
                     total_tokens=173,
-                    details={'thoughts_token_count': 101},
+                    details={'thoughts_tokens': 101, 'text_prompt_tokens': 57},
                 ),
                 model_name='models/gemini-2.5-pro-preview-05-06',
                 timestamp=IsDatetime(),
+                vendor_details={'finish_reason': 'STOP'},
             ),
             ModelRequest(
                 parts=[
@@ -232,9 +270,16 @@ async def test_google_model_retry(allow_model_requests: None, google_provider: G
                         content='I am sorry, I cannot fulfill this request. The country you provided is not supported.'
                     )
                 ],
-                usage=Usage(requests=1, request_tokens=104, response_tokens=18, total_tokens=122, details={}),
+                usage=Usage(
+                    requests=1,
+                    request_tokens=104,
+                    response_tokens=18,
+                    total_tokens=122,
+                    details={'text_prompt_tokens': 104},
+                ),
                 model_name='models/gemini-2.5-pro-preview-05-06',
                 timestamp=IsDatetime(),
+                vendor_details={'finish_reason': 'STOP'},
             ),
         ]
     )
@@ -488,9 +533,16 @@ async def test_google_model_instructions(allow_model_requests: None, google_prov
             ),
             ModelResponse(
                 parts=[TextPart(content='The capital of France is Paris.\n')],
-                usage=Usage(requests=1, request_tokens=13, response_tokens=8, total_tokens=21, details={}),
+                usage=Usage(
+                    requests=1,
+                    request_tokens=13,
+                    response_tokens=8,
+                    total_tokens=21,
+                    details={'text_prompt_tokens': 13, 'text_candidates_tokens': 8},
+                ),
                 model_name='gemini-2.0-flash',
                 timestamp=IsDatetime(),
+                vendor_details={'finish_reason': 'STOP'},
             ),
         ]
     )