Merge branch 'main' into add-googlemodel-google-specific-error-handling

Author: DouweM

.github/workflows/ci.yml

@@ -80,7 +80,7 @@ jobs:
       - run: make docs
 
       - run: make docs-insiders
-        if: github.event.pull_request.head.repo.full_name == github.repository || github.ref == 'refs/heads/main'
+        if: (github.event.pull_request.head.repo.full_name == github.repository || github.ref == 'refs/heads/main') && github.repository == 'pydantic/pydantic-ai'
         env:
           PPPR_TOKEN: ${{ secrets.PPPR_TOKEN }}
 
@@ -103,7 +103,7 @@ jobs:
   test-live:
     runs-on: ubuntu-latest
     timeout-minutes: 5
-    if: github.event.pull_request.head.repo.full_name == github.repository || github.event_name == 'push'
+    if: (github.event.pull_request.head.repo.full_name == github.repository || github.event_name == 'push') && github.repository == 'pydantic/pydantic-ai'
     steps:
       - uses: actions/checkout@v4
 
@@ -202,7 +202,8 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        python-version: ["3.10", "3.11", "3.12", "3.13"]
+        # TODO(Marcelo): Enable 3.11 again.
+        python-version: ["3.10", "3.12", "3.13"]
     env:
       CI: true
       COVERAGE_PROCESS_START: ./pyproject.toml

.gitignore

@@ -21,3 +21,5 @@ node_modules/
 /test_tmp/
 .mcp.json
 .claude/
+/.cursor/
+/.devcontainer/

Makefile

@@ -53,16 +53,16 @@ typecheck-both: typecheck-pyright typecheck-mypy
 .PHONY: test
 test: ## Run tests and collect coverage data
 	@# To test using a specific version of python, run 'make install-all-python' then set environment variable PYTEST_PYTHON=3.10 or similar
-	$(if $(PYTEST_PYTHON),UV_PROJECT_ENVIRONMENT=.venv$(subst .,,$(PYTEST_PYTHON))) uv run $(if $(PYTEST_PYTHON),--python $(PYTEST_PYTHON)) coverage run -m pytest -n auto --dist=loadgroup --durations=20
+	COLUMNS=150 $(if $(PYTEST_PYTHON),UV_PROJECT_ENVIRONMENT=.venv$(subst .,,$(PYTEST_PYTHON))) uv run $(if $(PYTEST_PYTHON),--python $(PYTEST_PYTHON)) coverage run -m pytest -n auto --dist=loadgroup --durations=20
 	@uv run coverage combine
 	@uv run coverage report
 
 .PHONY: test-all-python
 test-all-python: ## Run tests on Python 3.10 to 3.13
-	UV_PROJECT_ENVIRONMENT=.venv310 uv run --python 3.10 --all-extras --all-packages coverage run -p -m pytest
-	UV_PROJECT_ENVIRONMENT=.venv311 uv run --python 3.11 --all-extras --all-packages coverage run -p -m pytest
-	UV_PROJECT_ENVIRONMENT=.venv312 uv run --python 3.12 --all-extras --all-packages coverage run -p -m pytest
-	UV_PROJECT_ENVIRONMENT=.venv313 uv run --python 3.13 --all-extras --all-packages coverage run -p -m pytest
+	COLUMNS=150 UV_PROJECT_ENVIRONMENT=.venv310 uv run --python 3.10 --all-extras --all-packages coverage run -p -m pytest
+	COLUMNS=150 UV_PROJECT_ENVIRONMENT=.venv311 uv run --python 3.11 --all-extras --all-packages coverage run -p -m pytest
+	COLUMNS=150 UV_PROJECT_ENVIRONMENT=.venv312 uv run --python 3.12 --all-extras --all-packages coverage run -p -m pytest
+	COLUMNS=150 UV_PROJECT_ENVIRONMENT=.venv313 uv run --python 3.13 --all-extras --all-packages coverage run -p -m pytest
 	@uv run coverage combine
 	@uv run coverage report
 

docs/.hooks/main.py

@@ -16,10 +16,12 @@
 
 def on_page_markdown(markdown: str, page: Page, config: Config, files: Files) -> str:
     """Called on each file after it is read and before it is converted to HTML."""
-    markdown = inject_snippets(markdown, (DOCS_ROOT / page.file.src_uri).parent)
+    relative_path_root = (DOCS_ROOT / page.file.src_uri).parent
+    markdown = inject_snippets(markdown, relative_path_root)
     markdown = replace_uv_python_run(markdown)
     markdown = render_examples(markdown)
     markdown = render_video(markdown)
+    markdown = create_gateway_toggle(markdown, relative_path_root)
     return markdown
 
 
@@ -39,6 +41,7 @@ def on_env(env: Environment, config: Config, files: Files) -> Environment:
 
 def on_post_build(config: Config) -> None:
     """Inject extra CSS into mermaid styles to avoid titles being the same color as the background in dark mode."""
+    assert bundle_path is not None
     if bundle_path.exists():
         content = bundle_path.read_text()
         content, _ = re.subn(r'}(\.statediagram)', '}.statediagramTitleText{fill:#888}\1', content, count=1)
@@ -115,3 +118,109 @@ def sub_cf_video(m: re.Match[str]) -> str:
   ></iframe>
 </div>
 """
+
+
+def create_gateway_toggle(markdown: str, relative_path_root: Path) -> str:
+    """Transform Python code blocks with Agent() calls to show both Pydantic AI and Gateway versions."""
+    # Pattern matches Python code blocks with or without attributes, and optional annotation definitions after
+    # Annotation definitions are numbered list items like "1. Some text" that follow the code block
+    return re.sub(
+        r'```py(?:thon)?(?: *\{?([^}\n]*)\}?)?\n(.*?)\n```(\n\n(?:\d+\..+?\n)+?\n)?',
+        lambda m: transform_gateway_code_block(m, relative_path_root),
+        markdown,
+        flags=re.MULTILINE | re.DOTALL,
+    )
+
+
+# Models that should get gateway transformation
+GATEWAY_MODELS = ('anthropic', 'openai', 'openai-responses', 'openai-chat', 'bedrock', 'google-vertex', 'groq')
+
+
+def transform_gateway_code_block(m: re.Match[str], relative_path_root: Path) -> str:
+    """Transform a single code block to show both versions if it contains Agent() calls."""
+    attrs = m.group(1) or ''
+    code = m.group(2)
+    annotations = m.group(3) or ''  # Capture annotation definitions if present
+
+    # Simple check: does the code contain both "Agent(" and a quoted string?
+    if 'Agent(' not in code:
+        attrs_str = f' {{{attrs}}}' if attrs else ''
+        return f'```python{attrs_str}\n{code}\n```{annotations}'
+
+    # Check if code contains Agent() with a model that should be transformed
+    # Look for Agent(...'model:...' or Agent(..."model:..."
+    agent_pattern = r'Agent\((?:(?!["\']).)*([\"\'])([^"\']+)\1'
+    agent_match = re.search(agent_pattern, code, flags=re.DOTALL)
+
+    if not agent_match:
+        # No Agent() with string literal found
+        attrs_str = f' {{{attrs}}}' if attrs else ''
+        return f'```python{attrs_str}\n{code}\n```{annotations}'
+
+    model_string = agent_match.group(2)
+    # Check if model starts with one of the gateway-supported models
+    should_transform = any(model_string.startswith(f'{model}:') for model in GATEWAY_MODELS)
+
+    if not should_transform:
+        # Model doesn't match gateway models, return original
+        attrs_str = f' {{{attrs}}}' if attrs else ''
+        return f'```python{attrs_str}\n{code}\n```{annotations}'
+
+    # Transform the code for gateway version
+    def replace_agent_model(match: re.Match[str]) -> str:
+        """Replace model string with gateway/ prefix."""
+        full_match = match.group(0)
+        quote = match.group(1)
+        model = match.group(2)
+
+        # Replace the model string while preserving the rest
+        return full_match.replace(f'{quote}{model}{quote}', f'{quote}gateway/{model}{quote}', 1)
+
+    # This pattern finds: "Agent(" followed by anything (lazy), then the first quoted string
+    gateway_code = re.sub(
+        agent_pattern,
+        replace_agent_model,
+        code,
+        flags=re.DOTALL,
+    )
+
+    # Build attributes string
+    docs_path = DOCS_ROOT / 'gateway'
+    relative_path = docs_path.relative_to(relative_path_root, walk_up=True)
+    link = f"<a href='{relative_path}' style='float: right;'>Learn about Gateway</a>"
+
+    attrs_str = f' {{{attrs}}}' if attrs else ''
+
+    if 'title="' in attrs:
+        gateway_attrs = attrs.replace('title="', f'title="{link} ', 1)
+    else:
+        gateway_attrs = attrs + f' title="{link}"'
+    gateway_attrs_str = f' {{{gateway_attrs}}}'
+
+    # Indent code lines for proper markdown formatting within tabs
+    # Always add 4 spaces to every line (even empty ones) to preserve annotations
+    code_lines = code.split('\n')
+    indented_code = '\n'.join('    ' + line for line in code_lines)
+
+    gateway_code_lines = gateway_code.split('\n')
+    indented_gateway_code = '\n'.join('    ' + line for line in gateway_code_lines)
+
+    # Indent annotation definitions if present (need to be inside tabs for Material to work)
+    indented_annotations = ''
+    if annotations:
+        # Remove surrounding newlines and indent each line with 4 spaces
+        annotation_lines = annotations.strip().split('\n')
+        indented_annotations = '\n\n' + '\n'.join('    ' + line for line in annotation_lines) + '\n\n'
+
+    return f"""\
+=== "With Pydantic AI Gateway"
+
+    ```python{gateway_attrs_str}
+{indented_gateway_code}
+    ```{indented_annotations}
+
+=== "Directly to Provider API"
+
+    ```python{attrs_str}
+{indented_code}
+    ```{indented_annotations}"""

docs/.overrides/main.html

@@ -0,0 +1,8 @@
+{% extends "base.html" %}
+
+{% block announce %}
+    <strong>
+        <a href="/gateway">Pydantic AI Gateway</a> is now available! 🚀
+        Enterprise-ready AI model routing: One key for all your models with real-time monitoring and budget control that works.
+    </strong>
+{% endblock %}

docs/agents.md

@@ -320,7 +320,8 @@ async def main():
                         content='What is the capital of France?',
                         timestamp=datetime.datetime(...),
                     )
-                ]
+                ],
+                run_id='...',
             )
         ),
         CallToolsNode(
@@ -329,6 +330,7 @@ async def main():
                 usage=RequestUsage(input_tokens=56, output_tokens=7),
                 model_name='gpt-5',
                 timestamp=datetime.datetime(...),
+                run_id='...',
             )
         ),
         End(data=FinalResult(output='The capital of France is Paris.')),
@@ -382,7 +384,8 @@ async def main():
                             content='What is the capital of France?',
                             timestamp=datetime.datetime(...),
                         )
-                    ]
+                    ],
+                    run_id='...',
                 )
             ),
             CallToolsNode(
@@ -391,6 +394,7 @@ async def main():
                     usage=RequestUsage(input_tokens=56, output_tokens=7),
                     model_name='gpt-5',
                     timestamp=datetime.datetime(...),
+                    run_id='...',
                 )
             ),
             End(data=FinalResult(output='The capital of France is Paris.')),
@@ -1044,7 +1048,8 @@ with capture_run_messages() as messages:  # (2)!
                         content='Please get me the volume of a box with size 6.',
                         timestamp=datetime.datetime(...),
                     )
-                ]
+                ],
+                run_id='...',
             ),
             ModelResponse(
                 parts=[
@@ -1057,6 +1062,7 @@ with capture_run_messages() as messages:  # (2)!
                 usage=RequestUsage(input_tokens=62, output_tokens=4),
                 model_name='gpt-5',
                 timestamp=datetime.datetime(...),
+                run_id='...',
             ),
             ModelRequest(
                 parts=[
@@ -1066,7 +1072,8 @@ with capture_run_messages() as messages:  # (2)!
                         tool_call_id='pyd_ai_tool_call_id',
                         timestamp=datetime.datetime(...),
                     )
-                ]
+                ],
+                run_id='...',
             ),
             ModelResponse(
                 parts=[
@@ -1079,6 +1086,7 @@ with capture_run_messages() as messages:  # (2)!
                 usage=RequestUsage(input_tokens=72, output_tokens=8),
                 model_name='gpt-5',
                 timestamp=datetime.datetime(...),
+                run_id='...',
             ),
         ]
         """

docs/api/models/function.md

@@ -29,7 +29,8 @@ async def model_function(
                     content='Testing my agent...',
                     timestamp=datetime.datetime(...),
                 )
-            ]
+            ],
+            run_id='...',
         )
     ]
     """

docs/deferred-tools.md

@@ -106,7 +106,8 @@ print(result.all_messages())
                 content='Delete `__init__.py`, write `Hello, world!` to `README.md`, and clear `.env`',
                 timestamp=datetime.datetime(...),
             )
-        ]
+        ],
+        run_id='...',
     ),
     ModelResponse(
         parts=[
@@ -129,6 +130,7 @@ print(result.all_messages())
         usage=RequestUsage(input_tokens=63, output_tokens=21),
         model_name='gpt-5',
         timestamp=datetime.datetime(...),
+        run_id='...',
     ),
     ModelRequest(
         parts=[
@@ -138,23 +140,25 @@ print(result.all_messages())
                 tool_call_id='update_file_readme',
                 timestamp=datetime.datetime(...),
             )
-        ]
+        ],
+        run_id='...',
     ),
     ModelRequest(
         parts=[
-            ToolReturnPart(
-                tool_name='delete_file',
-                content='Deleting files is not allowed',
-                tool_call_id='delete_file',
-                timestamp=datetime.datetime(...),
-            ),
             ToolReturnPart(
                 tool_name='update_file',
                 content="File '.env' updated: ''",
                 tool_call_id='update_file_dotenv',
                 timestamp=datetime.datetime(...),
             ),
-        ]
+            ToolReturnPart(
+                tool_name='delete_file',
+                content='Deleting files is not allowed',
+                tool_call_id='delete_file',
+                timestamp=datetime.datetime(...),
+            ),
+        ],
+        run_id='...',
     ),
     ModelResponse(
         parts=[
@@ -165,6 +169,7 @@ print(result.all_messages())
         usage=RequestUsage(input_tokens=79, output_tokens=39),
         model_name='gpt-5',
         timestamp=datetime.datetime(...),
+        run_id='...',
     ),
 ]
 """
@@ -275,7 +280,8 @@ async def main():
                     content='Calculate the answer to the ultimate question of life, the universe, and everything',
                     timestamp=datetime.datetime(...),
                 )
-            ]
+            ],
+            run_id='...',
         ),
         ModelResponse(
             parts=[
@@ -290,6 +296,7 @@ async def main():
             usage=RequestUsage(input_tokens=63, output_tokens=13),
             model_name='gpt-5',
             timestamp=datetime.datetime(...),
+            run_id='...',
         ),
         ModelRequest(
             parts=[
@@ -299,7 +306,8 @@ async def main():
                     tool_call_id='pyd_ai_tool_call_id',
                     timestamp=datetime.datetime(...),
                 )
-            ]
+            ],
+            run_id='...',
         ),
         ModelResponse(
             parts=[
@@ -310,6 +318,7 @@ async def main():
             usage=RequestUsage(input_tokens=64, output_tokens=28),
             model_name='gpt-5',
             timestamp=datetime.datetime(...),
+            run_id='...',
         ),
     ]
     """

docs/durable_execution/temporal.md

@@ -172,7 +172,7 @@ As workflows and activities run in separate processes, any values passed between
 
 To account for these limitations, tool functions and the [event stream handler](#streaming) running inside activities receive a limited version of the agent's [`RunContext`][pydantic_ai.tools.RunContext], and it's your responsibility to make sure that the [dependencies](../dependencies.md) object provided to [`TemporalAgent.run()`][pydantic_ai.durable_exec.temporal.TemporalAgent.run] can be serialized using Pydantic.
 
-Specifically, only the `deps`, `retries`, `tool_call_id`, `tool_name`, `tool_call_approved`, `retry`, `max_retries`, `run_step` and `partial_output` fields are available by default, and trying to access `model`, `usage`, `prompt`, `messages`, or `tracer` will raise an error.
+Specifically, only the `deps`, `run_id`, `retries`, `tool_call_id`, `tool_name`, `tool_call_approved`, `retry`, `max_retries`, `run_step` and `partial_output` fields are available by default, and trying to access `model`, `usage`, `prompt`, `messages`, or `tracer` will raise an error.
 If you need one or more of these attributes to be available inside activities, you can create a [`TemporalRunContext`][pydantic_ai.durable_exec.temporal.TemporalRunContext] subclass with custom `serialize_run_context` and `deserialize_run_context` class methods and pass it to [`TemporalAgent`][pydantic_ai.durable_exec.temporal.TemporalAgent] as `run_context_type`.
 
 ### Streaming