pydantic-ai-slim 0.4.0__py3-none-any.whl → 0.4.1__py3-none-any.whl

pydantic_ai/_a2a.py

@@ -1,11 +1,13 @@
 from __future__ import annotations, annotations as _annotations
 
+import uuid
 from collections.abc import AsyncIterator, Sequence
 from contextlib import asynccontextmanager
 from dataclasses import dataclass
 from functools import partial
-from typing import Any, Generic
+from typing import Any, Generic, TypeVar
 
+from pydantic import TypeAdapter
 from typing_extensions import assert_never
 
 from pydantic_ai.messages import (
@@ -19,12 +21,17 @@ from pydantic_ai.messages import (
     ModelResponse,
     ModelResponsePart,
     TextPart,
+    ThinkingPart,
+    ToolCallPart,
     UserPromptPart,
     VideoUrl,
 )
 
 from .agent import Agent, AgentDepsT, OutputDataT
 
+# AgentWorker output type needs to be invariant for use in both parameter and return positions
+WorkerOutputT = TypeVar('WorkerOutputT')
+
 try:
     from starlette.middleware import Middleware
     from starlette.routing import Route
@@ -33,10 +40,11 @@ try:
     from fasta2a.applications import FastA2A
     from fasta2a.broker import Broker, InMemoryBroker
     from fasta2a.schema import (
+        AgentProvider,
         Artifact,
+        DataPart,
         Message,
         Part,
-        Provider,
         Skill,
         TaskIdParams,
         TaskSendParams,
@@ -72,7 +80,7 @@ def agent_to_a2a(
     url: str = 'http://localhost:8000',
     version: str = '1.0.0',
     description: str | None = None,
-    provider: Provider | None = None,
+    provider: AgentProvider | None = None,
     skills: list[Skill] | None = None,
     # Starlette
     debug: bool = False,
@@ -106,59 +114,121 @@ def agent_to_a2a(
 
 
 @dataclass
-class AgentWorker(Worker, Generic[AgentDepsT, OutputDataT]):
+class AgentWorker(Worker[list[ModelMessage]], Generic[WorkerOutputT, AgentDepsT]):
     """A worker that uses an agent to execute tasks."""
 
-    agent: Agent[AgentDepsT, OutputDataT]
+    agent: Agent[AgentDepsT, WorkerOutputT]
 
     async def run_task(self, params: TaskSendParams) -> None:
-        task = await self.storage.load_task(params['id'], history_length=params.get('history_length'))
-        assert task is not None, f'Task {params["id"]} not found'
-        assert 'session_id' in task, 'Task must have a session_id'
+        task = await self.storage.load_task(params['id'])
+        if task is None:
+            raise ValueError(f'Task {params["id"]} not found')  # pragma: no cover
+
+        # TODO(Marcelo): Should we lock `run_task` on the `context_id`?
+        # Ensure this task hasn't been run before
+        if task['status']['state'] != 'submitted':
+            raise ValueError(  # pragma: no cover
+                f'Task {params["id"]} has already been processed (state: {task["status"]["state"]})'
+            )
 
         await self.storage.update_task(task['id'], state='working')
 
-        # TODO(Marcelo): We need to have a way to communicate when the task is set to `input-required`. Maybe
-        # a custom `output_type` with a `more_info_required` field, or something like that.
+        # Load context - contains pydantic-ai message history from previous tasks in this conversation
+        message_history = await self.storage.load_context(task['context_id']) or []
+        message_history.extend(self.build_message_history(task.get('history', [])))
+
+        try:
+            result = await self.agent.run(message_history=message_history)  # type: ignore
 
-        task_history = task.get('history', [])
-        message_history = self.build_message_history(task_history=task_history)
+            await self.storage.update_context(task['context_id'], result.all_messages())
 
-        # TODO(Marcelo): We need to make this more customizable e.g. pass deps.
-        result = await self.agent.run(message_history=message_history)  # type: ignore
+            # Convert new messages to A2A format for task history
+            a2a_messages: list[Message] = []
+
+            for message in result.new_messages():
+                if isinstance(message, ModelRequest):
+                    # Skip user prompts - they're already in task history
+                    continue
+                else:
+                    # Convert response parts to A2A format
+                    a2a_parts = self._response_parts_to_a2a(message.parts)
+                    if a2a_parts:  # Add if there are visible parts (text/thinking)
+                        a2a_messages.append(
+                            Message(role='agent', parts=a2a_parts, kind='message', message_id=str(uuid.uuid4()))
+                        )
 
-        artifacts = self.build_artifacts(result.output)
-        await self.storage.update_task(task['id'], state='completed', artifacts=artifacts)
+            artifacts = self.build_artifacts(result.output)
+        except Exception:
+            await self.storage.update_task(task['id'], state='failed')
+            raise
+        else:
+            await self.storage.update_task(
+                task['id'], state='completed', new_artifacts=artifacts, new_messages=a2a_messages
+            )
 
     async def cancel_task(self, params: TaskIdParams) -> None:
         pass
 
-    def build_artifacts(self, result: Any) -> list[Artifact]:
-        # TODO(Marcelo): We need to send the json schema of the result on the metadata of the message.
-        return [Artifact(name='result', index=0, parts=[A2ATextPart(type='text', text=str(result))])]
+    def build_artifacts(self, result: WorkerOutputT) -> list[Artifact]:
+        """Build artifacts from agent result.
 
-    def build_message_history(self, task_history: list[Message]) -> list[ModelMessage]:
+        All agent outputs become artifacts to mark them as durable task outputs.
+        For string results, we use TextPart. For structured data, we use DataPart.
+        Metadata is included to preserve type information.
+        """
+        artifact_id = str(uuid.uuid4())
+        part = self._convert_result_to_part(result)
+        return [Artifact(artifact_id=artifact_id, name='result', parts=[part])]
+
+    def _convert_result_to_part(self, result: WorkerOutputT) -> Part:
+        """Convert agent result to a Part (TextPart or DataPart).
+
+        For string results, returns a TextPart.
+        For structured data, returns a DataPart with properly serialized data.
+        """
+        if isinstance(result, str):
+            return A2ATextPart(kind='text', text=result)
+        else:
+            output_type = type(result)
+            type_adapter = TypeAdapter(output_type)
+            data = type_adapter.dump_python(result, mode='json')
+            json_schema = type_adapter.json_schema(mode='serialization')
+            return DataPart(kind='data', data={'result': data}, metadata={'json_schema': json_schema})
+
+    def build_message_history(self, history: list[Message]) -> list[ModelMessage]:
         model_messages: list[ModelMessage] = []
-        for message in task_history:
+        for message in history:
             if message['role'] == 'user':
-                model_messages.append(ModelRequest(parts=self._map_request_parts(message['parts'])))
+                model_messages.append(ModelRequest(parts=self._request_parts_from_a2a(message['parts'])))
             else:
-                model_messages.append(ModelResponse(parts=self._map_response_parts(message['parts'])))
+                model_messages.append(ModelResponse(parts=self._response_parts_from_a2a(message['parts'])))
         return model_messages
 
-    def _map_request_parts(self, parts: list[Part]) -> list[ModelRequestPart]:
+    def _request_parts_from_a2a(self, parts: list[Part]) -> list[ModelRequestPart]:
+        """Convert A2A Part objects to pydantic-ai ModelRequestPart objects.
+
+        This handles the conversion from A2A protocol parts (text, file, data) to
+        pydantic-ai's internal request parts (UserPromptPart with various content types).
+
+        Args:
+            parts: List of A2A Part objects from incoming messages
+
+        Returns:
+            List of ModelRequestPart objects for the pydantic-ai agent
+        """
         model_parts: list[ModelRequestPart] = []
         for part in parts:
-            if part['type'] == 'text':
+            if part['kind'] == 'text':
                 model_parts.append(UserPromptPart(content=part['text']))
-            elif part['type'] == 'file':
-                file = part['file']
-                if 'data' in file:
-                    data = file['data'].encode('utf-8')
-                    content = BinaryContent(data=data, media_type=file['mime_type'])
+            elif part['kind'] == 'file':
+                file_content = part['file']
+                if 'bytes' in file_content:
+                    data = file_content['bytes'].encode('utf-8')
+                    mime_type = file_content.get('mime_type', 'application/octet-stream')
+                    content = BinaryContent(data=data, media_type=mime_type)
                     model_parts.append(UserPromptPart(content=[content]))
                 else:
-                    url = file['url']
+                    url = file_content['uri']
                     for url_cls in (DocumentUrl, AudioUrl, ImageUrl, VideoUrl):
                         content = url_cls(url=url)
                         try:
@@ -168,24 +238,68 @@ class AgentWorker(Worker, Generic[AgentDepsT, OutputDataT]):
                         else:
                             break
                     else:
-                        raise ValueError(f'Unknown file type: {file["mime_type"]}')  # pragma: no cover
+                        raise ValueError(f'Unsupported file type: {url}')  # pragma: no cover
                     model_parts.append(UserPromptPart(content=[content]))
-            elif part['type'] == 'data':
-                # TODO(Marcelo): Maybe we should use this for `ToolReturnPart`, and `RetryPromptPart`.
+            elif part['kind'] == 'data':
                 raise NotImplementedError('Data parts are not supported yet.')
             else:
                 assert_never(part)
         return model_parts
 
-    def _map_response_parts(self, parts: list[Part]) -> list[ModelResponsePart]:
+    def _response_parts_from_a2a(self, parts: list[Part]) -> list[ModelResponsePart]:
+        """Convert A2A Part objects to pydantic-ai ModelResponsePart objects.
+
+        This handles the conversion from A2A protocol parts (text, file, data) to
+        pydantic-ai's internal response parts. Currently only supports text parts
+        as agent responses in A2A are expected to be text-based.
+
+        Args:
+            parts: List of A2A Part objects from stored agent messages
+
+        Returns:
+            List of ModelResponsePart objects for message history
+        """
         model_parts: list[ModelResponsePart] = []
         for part in parts:
-            if part['type'] == 'text':
+            if part['kind'] == 'text':
                 model_parts.append(TextPart(content=part['text']))
-            elif part['type'] == 'file':  # pragma: no cover
+            elif part['kind'] == 'file':  # pragma: no cover
                 raise NotImplementedError('File parts are not supported yet.')
-            elif part['type'] == 'data':  # pragma: no cover
+            elif part['kind'] == 'data':  # pragma: no cover
                 raise NotImplementedError('Data parts are not supported yet.')
             else:  # pragma: no cover
                 assert_never(part)
         return model_parts
+
+    def _response_parts_to_a2a(self, parts: list[ModelResponsePart]) -> list[Part]:
+        """Convert pydantic-ai ModelResponsePart objects to A2A Part objects.
+
+        This handles the conversion from pydantic-ai's internal response parts to
+        A2A protocol parts. Different part types are handled as follows:
+        - TextPart: Converted directly to A2A TextPart
+        - ThinkingPart: Converted to TextPart with metadata indicating it's thinking
+        - ToolCallPart: Skipped (internal to agent execution)
+
+        Args:
+            parts: List of ModelResponsePart objects from agent response
+
+        Returns:
+            List of A2A Part objects suitable for sending via A2A protocol
+        """
+        a2a_parts: list[Part] = []
+        for part in parts:
+            if isinstance(part, TextPart):
+                a2a_parts.append(A2ATextPart(kind='text', text=part.content))
+            elif isinstance(part, ThinkingPart):
+                # Convert thinking to text with metadata
+                a2a_parts.append(
+                    A2ATextPart(
+                        kind='text',
+                        text=part.content,
+                        metadata={'type': 'thinking', 'thinking_id': part.id, 'signature': part.signature},
+                    )
+                )
+            elif isinstance(part, ToolCallPart):
+                # Skip tool calls - they're internal to agent execution
+                pass
+        return a2a_parts

pydantic_ai/agent.py

@@ -63,7 +63,7 @@ if TYPE_CHECKING:
 
     from fasta2a.applications import FastA2A
     from fasta2a.broker import Broker
-    from fasta2a.schema import Provider, Skill
+    from fasta2a.schema import AgentProvider, Skill
     from fasta2a.storage import Storage
     from pydantic_ai.mcp import MCPServer
 
@@ -1764,7 +1764,7 @@ class Agent(Generic[AgentDepsT, OutputDataT]):
         url: str = 'http://localhost:8000',
         version: str = '1.0.0',
         description: str | None = None,
-        provider: Provider | None = None,
+        provider: AgentProvider | None = None,
         skills: list[Skill] | None = None,
         # Starlette
         debug: bool = False,

{pydantic_ai_slim-0.4.0.dist-info → pydantic_ai_slim-0.4.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pydantic-ai-slim
-Version: 0.4.0
+Version: 0.4.1
 Summary: Agent Framework / shim to use Pydantic with LLMs, slim package
 Author-email: Samuel Colvin <samuel@pydantic.dev>, Marcelo Trylesinski <marcelotryle@gmail.com>, David Montague <david@pydantic.dev>, Alex Hall <alex@pydantic.dev>
 License-Expression: MIT
@@ -30,11 +30,11 @@ Requires-Dist: exceptiongroup; python_version < '3.11'
 Requires-Dist: griffe>=1.3.2
 Requires-Dist: httpx>=0.27
 Requires-Dist: opentelemetry-api>=1.28.0
-Requires-Dist: pydantic-graph==0.4.0
+Requires-Dist: pydantic-graph==0.4.1
 Requires-Dist: pydantic>=2.10
 Requires-Dist: typing-inspection>=0.4.0
 Provides-Extra: a2a
-Requires-Dist: fasta2a==0.4.0; extra == 'a2a'
+Requires-Dist: fasta2a==0.4.1; extra == 'a2a'
 Provides-Extra: anthropic
 Requires-Dist: anthropic>=0.52.0; extra == 'anthropic'
 Provides-Extra: bedrock
@@ -48,7 +48,7 @@ Requires-Dist: cohere>=5.13.11; (platform_system != 'Emscripten') and extra == '
 Provides-Extra: duckduckgo
 Requires-Dist: duckduckgo-search>=7.0.0; extra == 'duckduckgo'
 Provides-Extra: evals
-Requires-Dist: pydantic-evals==0.4.0; extra == 'evals'
+Requires-Dist: pydantic-evals==0.4.1; extra == 'evals'
 Provides-Extra: google
 Requires-Dist: google-genai>=1.24.0; extra == 'google'
 Provides-Extra: groq

{pydantic_ai_slim-0.4.0.dist-info → pydantic_ai_slim-0.4.1.dist-info}/RECORD

@@ -1,6 +1,6 @@
 pydantic_ai/__init__.py,sha256=Ns04g4Efqkzwccs8w2nGphfWbptMlIJYG8vIJbGGyG0,1262
 pydantic_ai/__main__.py,sha256=Q_zJU15DUA01YtlJ2mnaLCoId2YmgmreVEERGuQT-Y0,132
-pydantic_ai/_a2a.py,sha256=8nNtx6GENDt2Ej3f1ui9L-FuNQBYVELpJFfwz-y7fUw,7234
+pydantic_ai/_a2a.py,sha256=G6W8zLRE5FNug19GieVkYuGPw5CA44YeZnS7GTN7M30,12068
 pydantic_ai/_agent_graph.py,sha256=rtzyBXN4bzEDBeRkRwF031ORktSMbuGz9toZmSqUxNI,42153
 pydantic_ai/_cli.py,sha256=R-sE-9gYqPxV5-5utso4g-bzAKMiTCdo33XOVqE0ZEg,13206
 pydantic_ai/_function_schema.py,sha256=BZus5y51eqiGQKxQIcCiDoSPml3AtAb12-st_aujU2k,10813
@@ -12,7 +12,7 @@ pydantic_ai/_run_context.py,sha256=zNkSyiQSH-YweO39ii3iB2taouUOodo3sTjz2Lrj4Pc,1
 pydantic_ai/_system_prompt.py,sha256=lUSq-gDZjlYTGtd6BUm54yEvTIvgdwBmJ8mLsNZZtYU,1142
 pydantic_ai/_thinking_part.py,sha256=mzx2RZSfiQxAKpljEflrcXRXmFKxtp6bKVyorY3UYZk,1554
 pydantic_ai/_utils.py,sha256=SGXEiGCnMae1Iz_eZKUs6ni_tGMPkDaJ4W3W3YMoP5w,15545
-pydantic_ai/agent.py,sha256=Fs-bm9eeCvanwiKiD-IS_XLcMmgNWucJylXgrIDH6WM,96186
+pydantic_ai/agent.py,sha256=zvQgEG9eFG7entCTum3QSApHbNU8RvAE_ydscPaMAC4,96196
 pydantic_ai/direct.py,sha256=WRfgke3zm-eeR39LTuh9XI2TrdHXAqO81eDvFwih4Ko,14803
 pydantic_ai/exceptions.py,sha256=IdFw594Ou7Vn4YFa7xdZ040_j_6nmyA3MPANbC7sys4,3175
 pydantic_ai/format_as_xml.py,sha256=IINfh1evWDphGahqHNLBArB5dQ4NIqS3S-kru35ztGg,372
@@ -76,8 +76,8 @@ pydantic_ai/providers/mistral.py,sha256=EIUSENjFuGzBhvbdrarUTM4VPkesIMnZrzfnEKHO
 pydantic_ai/providers/openai.py,sha256=7iGij0EaFylab7dTZAZDgXr78tr-HsZrn9EI9AkWBNQ,3091
 pydantic_ai/providers/openrouter.py,sha256=NXjNdnlXIBrBMMqbzcWQnowXOuZh4NHikXenBn5h3mc,4061
 pydantic_ai/providers/together.py,sha256=zFVSMSm5jXbpkNouvBOTjWrPmlPpCp6sQS5LMSyVjrQ,3482
-pydantic_ai_slim-0.4.0.dist-info/METADATA,sha256=S-ygqOZ0lpsazK_VGyrj8B6l1H9Q7B2bCGYRtmmK4T8,3846
-pydantic_ai_slim-0.4.0.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-pydantic_ai_slim-0.4.0.dist-info/entry_points.txt,sha256=kbKxe2VtDCYS06hsI7P3uZGxcVC08-FPt1rxeiMpIps,50
-pydantic_ai_slim-0.4.0.dist-info/licenses/LICENSE,sha256=vA6Jc482lEyBBuGUfD1pYx-cM7jxvLYOxPidZ30t_PQ,1100
-pydantic_ai_slim-0.4.0.dist-info/RECORD,,
+pydantic_ai_slim-0.4.1.dist-info/METADATA,sha256=PqGrAd6qbv0rxMgiCa6N1Lo1mBvMb3XUC_FGaxuzeAY,3846
+pydantic_ai_slim-0.4.1.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+pydantic_ai_slim-0.4.1.dist-info/entry_points.txt,sha256=kbKxe2VtDCYS06hsI7P3uZGxcVC08-FPt1rxeiMpIps,50
+pydantic_ai_slim-0.4.1.dist-info/licenses/LICENSE,sha256=vA6Jc482lEyBBuGUfD1pYx-cM7jxvLYOxPidZ30t_PQ,1100
+pydantic_ai_slim-0.4.1.dist-info/RECORD,,