v0.3.6 (2025-07-04)

What's Changed

• Deprecate {FunctionToolCallEvent,FunctionToolResultEvent}.call_id in favor of tool_call_id by @proever in #2028
• Indicate to the model that a RetryPromptPart not tied to a tool call contains validation feedback rather than a user message by @hovi in #2008
• Update starlette subdomain in docs by @alDuncanson in #2099
• Update client.md - Typo by @kauabh in #2105
• Add support for predicted outputs in OpenAIModelSettings by @webholics in #2106
• Record tool response in tool run span by @alexmojaki in #2109
• Use contextvars for agent overriding, rather than a local attribute by @dmontagu in #2118
• Fix model parameters not being customized in fallback model request stream by @almeidaalajoel in #2120
• simplify weather example by @samuelcolvin in #2129

New Contributors

• @alDuncanson made their first contribution in #2099
• @webholics made their first contribution in #2106
• @almeidaalajoel made their first contribution in #2120

Full Changelog: v0.3.5...v0.3.6

v0.3.5 (2025-06-30)

What's Changed

• Add progress bar on evaluate by @davide-andreoli in #1871
• Fix deprecation warning under Pydantic 2.11 by @medaminezghal in #2076
• fix: async fixtures in conftest.py by @stevenh in #2068
• fix: docs examples python version in tests by @stevenh in #2069
• Set 'us-central1' by default on GoogleProvider by @Kludex in #2031
• Move ThinkingPart to preceed TextPart in OpenAIResponsesModel by @Kludex in #2043
• Fix deprecated kwargs validation to prevent silent failures by @svilupp in #2047
• Support strict mode in NativeOutput by @severinh in #2084
• Let tools return ToolReturn to pass additional content to model, or attach metadata that's not passed to the model by @Wh1isper in #2060
• Add ability to include snippets in docs with inline-named sections for fragments and highlighting by @dmontagu in #2088
• Add Slack Lead Qualifier example by @DouweM in #2079

New Contributors

• @medaminezghal made their first contribution in #2076
• @svilupp made their first contribution in #2047
• @severinh made their first contribution in #2084

Full Changelog: v0.3.4...v0.3.5

v0.3.4 (2025-06-26)

What's Changed

• Scrubbing sensitive content by @adtyavrdhn in #2014
• fix(anthropic): send ThinkingPart back when using tool calls by @Kludex in #2072

New Contributors

• @adtyavrdhn made their first contribution in #2014

Full Changelog: v0.3.3...v0.3.4

v0.3.3 (2025-06-24)

What's Changed

• Support NativeOutput and PromptedOutput modes in addition to ToolOutput by @DouweM in #1628
• Improve Dynamic Instructions Documentation by @xflashxx in #1819
• docs: add API reference to UserPromptNode by @Kludex in #2052
• Include additional usage fields from OpenAI-compatible APIs in usage details by @Wh1isper in #2038
• Fix: Handled and pretty-printed exceptions without exiting the CLI. by @fswair in #2055
• document alternative otel backends by @samuelcolvin in #2062
• note on self-hosting logfire by @samuelcolvin in #2063
• Make Edge hashable by @dmontagu in #2064

New Contributors

• @xflashxx made their first contribution in #1819

Full Changelog: v0.3.2...v0.3.3

v0.3.2 (2025-06-21)

What's Changed

• Support MCP sampling by @samuelcolvin in #1884

Full Changelog: v0.3.1...v0.3.2

v0.3.1 (2025-06-18)

What's Changed

• Update Google models by @tacoo in #2010
• fix: update ThinkingPart when delta contains signature by @Kludex in #2012

New Contributors

• @tacoo made their first contribution in #2010
• @a-klos made their first contribution in #2013

Full Changelog: v0.3.0...v0.3.1

v0.3.0 (2025-06-18)

Breaking Changes

See #1142 — Adds support for thinking parts.

We now convert the thinking blocks ("<think>..."</think>") in provider specific text parts to
PydanticAI ThinkingParts. Also, as part of this release, we made the choice to not send back the
ThinkingParts to the provider - the idea is to save costs on behalf of the user. In the future, we
intend to add a setting to customize this behavior.

---

What's Changed

• Support Thinking part by @Kludex in #1142

Full Changelog: v0.2.20...v0.3.0

v0.2.20 (2025-06-18)

What's Changed

• Handle McpError from MCP tool calls by @ppcantidio in #1999
• Add process_tool_call hook to MCP servers to modify tool args, metadata, and return value by @stevenh in #2000
• Respect ModelSettings.timeout in GoogleModel by @DouweM in #2006
• feat: add RunContext support to history processors by @Wh1isper in #2002

New Contributors

• @ppcantidio made their first contribution in #1999
• @stevenh made their first contribution in #2000

Full Changelog: v0.2.19...v0.2.20

v0.2.19 (2025-06-16)

What's Changed

• Proper check if callable is async by @Kludex in #1972
• More flexible method infer_provider by @hovi in #1945
• Ignore dynamic instructions returning an empty string by @giacbrd in #1961
• uprev Pyodide to 0.27.6 by @samuelcolvin in #1944
• refactor: updated tools doc with function naming: roll_die → roll_dice by @yamanahlawat in #1986
• Set Anthropic max_tokens to 4096 by default by @Kludex in #1994
• feat: add history_processors parameter to Agent for message processing by @Kludex in #1970
• Yield events for unknown tool calls by @proever in #1960
• Always set a parameters schema on a Gemini function declaration, even when it's an empty object by @DouweM in #1998

New Contributors

• @hovi made their first contribution in #1945
• @giacbrd made their first contribution in #1961
• @yamanahlawat made their first contribution in #1986
• @proever made their first contribution in #1960

Full Changelog: v0.2.18...v0.2.19

v0.2.18 (2025-06-13)

What's Changed

• Reuse last request from message history if no user prompt was provided by @DouweM in #1955
• Prevent Anthropic API errors from empty message content by @mike-luabase in #1934
• Add MCP Streamable HTTP implementation by @Kludex in #1965
• feat(openai): expose Responses API id as vendor_id by @sarunas-zebra in #1949
• Use GoogleModel instead of GeminiModel on inference by @Kludex in #1881

New Contributors

• @sarunas-zebra made their first contribution in #1949

Full Changelog: v0.2.17...v0.2.18