v1.0.13 (2025-10-01)

What's Changed

• feat: Otel instrumentation version 3 by @bitnahian in #3021
• Don't send strict to HuggingFace API as it's not supported by their types and at least some models by @DouweM in #3062
• Correctly merge Model.settings with model_settings in direct mode by @moritzwilksch in #2980
• Expose server_info in MCPServer by @Whadup in #3055
• Support contextually overriding agent instructions by @mwildehahn in #2926
• Update evals attributes by @dmontagu in #2924

New Contributors

• @Whadup made their first contribution in #3055

Full Changelog: v1.0.12...v1.0.13

v1.0.12 (2025-09-30)

What's Changed

• Support Anthropic built-in memory tool by @DouweM in #3042
• Support text, JSON, XML and YAML DocumentUrl and BinaryContent on OpenAI by @pulphix in #2851
• Prefer structuredContent MCP tool result when present by @shaheerzaman in #2854
• Expose .messages, .toolsets types in top-level pydantic_ai to aid IDE auto-import by @moritzwilksch in #2957
• Add cost metric to the pydantic-evals output by @dmontagu in #2984
• Add retry args to pydantic_evals.Dataset.evaluate_sync by @m7mdhka in #3022
• Change type of common_tools to work with agent with any deps type by @lukekh in #3037
• Don't error when Gemini returns more than one candidate response by @DouweM in #3043
• Handle Ollama responses without finish_reason and document Ollama Cloud by @DouweM in #3045
• Add custom Vertex AI Model Garden example by @stefannae in #2868
• Fix StructuredDict with nested JSON schemas using $ref by @ChiaXinLiang in #2570

New Contributors

• @m7mdhka made their first contribution in #3022
• @lukekh made their first contribution in #3037
• @pulphix made their first contribution in #2851
• @stefannae made their first contribution in #2868
• @ChiaXinLiang made their first contribution in #2570

Full Changelog: v1.0.11...v1.0.12

v1.0.11 (2025-09-29)

What's Changed

• Support OpenAI image detail on ImageUrl and BinaryContent via vendor_metadata by @moritzwilksch in #2987
• Support callable classes as history processors by @bitnahian in #2988
• Add claude-sonnet-4-5 to known model names by @DouweM in #3033
• Add operation.cost metric to instrumented models by @alexmojaki in #3013
• Fix streaming gpt-oss using Ollama by @DouweM in #3035
• Raise error when using Anthropic thinking with output tools by @DouweM in #3025
• Make OutputObjectDefinition public on pydantic_ai.output by @moritzwilksch in #2991
• Update pyproject.toml to be PEP639 compliant by @Kludex in #3001
• Bump temporalio to 1.18.0 by @DouweM in #3027
• Bump genai-prices to 0.0.28 by @DouweM in #3030
• Document that Gemini requires native or prompted output mode for structured output streaming by @DouweM in #3032
• Update Ollama docs instructions by @slkoo-cc in #2993
• Update docs and tests for DBOS v2.0 by @qianl15 in #3004

New Contributors

• @slkoo-cc made their first contribution in #2993

Full Changelog: v1.0.10...v1.0.11

v1.0.10 (2025-09-19)

What's Changed

• Fix OTel for built-in tools returning a list (e.g. Anthropic web search) by @DouweM in #2959
• Drop assertion that Google streaming chunk has candidates by @DouweM in #2960
• Retry model request that produced an empty response by @DouweM in #2961
• Clarify Agent(retries=...) description by @DouweM in #2963
• Stop redundantly encoding binary data as base64 when sending to Google genai SDK by @DouweM in #2962
• Use model class names as tags in format_as_xml and add option to include field titles and descriptions as attributes by @giacbrd in #2313

Full Changelog: v1.0.9...v1.0.10

v1.0.9 (2025-09-18)

What's Changed

• Stream built-in tool calls from OpenAI, Google, Anthropic and return them on next request (required for OpenAI reasoning) by @DouweM in #2877
• Include built-in tool calls and results in OTel messages by @DouweM in #2954
• Add RunContext.max_retries and .last_attempt by @DouweM in #2952
• Fix StreamedResponse.model_name for Azure OpenAI with content filter by @DouweM in #2951
• Fix TemporalAgent dropping model-specific ModelSettings (e.g. openai_reasoning_effort) by @DouweM in #2938
• Don't send item IDs to Responses API for non-reasoning models by @DouweM in #2950
• Update DBOS version by @qianl15 in #2939

Full Changelog: v1.0.8...v1.0.9

v1.0.8 (2025-09-16)

What's Changed

• Tools can now return AG-UI events separate from result sent to model by @DouweM in #2922
• Fix bug causing doubled reasoning tokens usage by deepcopying by @DouweM in #2920
• Fix auto-detection of HTTP proxy settings by @maxnilz in #2917
• Fix new_messages() and capture_run_messages() when history processors are used by @DouweM in #2921
• chore: Remove 'text' from RunUsage docstrings by @alexmojaki in #2919

New Contributors

• @maxnilz made their first contribution in #2917

Full Changelog: v1.0.7...v1.0.8

v1.0.7 (2025-09-15)

What's Changed

• Added MCP metadata and annotations to ToolDefinition.metadata for use in filtering by @ChuckJonas in #2880
• When starting run with message history ending in ModelRequest, make its content available in RunContext.prompt by @DouweM in #2891
• Let FunctionToolset take default values for strict, sequential, requires_approval, metadata by @DouweM in #2909
• Don't require mcp or logfire to use Temporal or DBOS by @DouweM in #2908
• Combine consecutive AG-UI user and assistant messages into the same model request/response by @DouweM in #2912
• Fix new_messages() when deferred_tool_results is used with message_history ending in ToolReturnParts by @DouweM in #2913

Full Changelog: v1.0.6...v1.0.7

v1.0.6 (2025-09-12)

What's Changed

• Add support for previous_response_id from Responses API by @GDaamn in #2756
• Let MCP servers be loaded from file by @Kludex in #2698
• Fix how thinking summaries are sent back to Responses API by @DouweM in #2883
• Bump Cohere SDK and remove incorrect typing workaround by @DouweM in #2886
• Update MCP tool call customisation docs by @MasterOdin in #2817

New Contributors

• @MasterOdin made their first contribution in #2817

Full Changelog: v1.0.5...v1.0.6

v1.0.5 (2025-09-11)

What's Changed

• Don't lose Azure OpenAI Responses encrypted_content if no summary was included by @DouweM in #2874
• Store OpenAI Responses text part ID to prevent error with reasoning by @DouweM in #2882
• Make OpenAIResponsesModel work with reasoning from other models and modified history by @DouweM in #2881

Full Changelog: v1.0.4...v1.0.5

v1.0.4 (2025-09-11)

What's Changed

• Add Pydantic AI Gateway provider by @Kludex in #2816, #2863
• Fix OpenAI Responses API tool calls with reasoning by @DouweM in #2869
• Support OpenAI Responses API returning encrypted reasoning content without summary by @DouweM in #2866
• Don't ask for OpenAI Responses API to include encrypted reasoning content for models that don't support it by @DouweM in #2867
• docs: update builtin-tools md by @tberends in #2857

New Contributors

• @tberends made their first contribution in #2857

Full Changelog: v1.0.3...v1.0.4