pydantic-ai Releases

Includes a security fix for Host header validation and introduces new models, web search capabilities, and configuration options.

Aug 14, 2026

Enhances MCP integration with support for newer SDK versions and adds Azure AI Voice Live capabilities.

Aug 13, 2026

Backports security fixes from v2.28.0 regarding Content-Type validation and retry prompt redaction, plus a build-tooling resolution note.

Aug 12, 2026

Backports security fixes from v2.28.0 regarding Content-Type validation and retry prompt redaction.

Aug 12, 2026

A major feature release adding realtime speech-to-speech support, new providers, and a security fix for Content-Type validation.

Aug 12, 2026

Bug fixes focusing on tool span restoration, streaming response mapping, and OpenTelemetry content gating.

Aug 11, 2026

Adds new provider support and settings, along with fixes for compaction handling and payload size limits.

Aug 8, 2026

Security patch to prevent unbounded memory use when downloading remote content via the local web_fetch tool.

Aug 8, 2026

Introduces first-party run cancellation capabilities, tool deferral features, and makes stream event iterators public for better agent lifecycle management.

Aug 7, 2026

Enhances xAI integration, fixes provider-specific settings for Azure and OpenAI models, and corrects thinking levels for specific GPT models.

Aug 6, 2026

A comprehensive bug fix release improving provider compatibility, specifically addressing Google provider timeouts, OpenRouter model name recognition, and Bedrock parameter handling.

Aug 5, 2026

Adds cost tracking functionality, restores stable model aliases, and improves Bedrock support by adding extra headers and tool availability rendering.

Aug 4, 2026

Provides a CUDA 12.9 wheel for the LMCache library release candidate to support GPU-accelerated caching.

Aug 1, 2026

Pre-release v0.5.3 of LMCache built for AMD ROCm 7.2, targeting new Instinct GPU models like MI300X and MI350X.

Aug 1, 2026

Enhances MCP tool handling, Anthropic system prompt integration, and adds `RunContext.is_tool_available`, alongside extensive Temporal and Prefect bug fixes.

Aug 1, 2026

Adds in-process Code Mode execution, a V8-isolated transport for secure execution, a built-in web search tool, and draft auto-publishing features.

Jul 31, 2026

Introduces chat channel support for agents, enhanced MongoDB vector store capabilities, exact metadata filtering, and performance improvements for streaming and sandbox execution.

Jul 30, 2026

This release adds a new token limit configuration for usage tracking and refreshes the list of known model names to ensure compatibility with Gateway probes and Google image IDs.

Jul 30, 2026

Adds read-only sandbox options for safer local execution, custom thread ID resolution, explicit channel reaction tools, and auto-publishing for agent updates.

Jul 29, 2026

Adds support for Claude Opus 5 and OpenAI Responses API reasoning context, while fixing bugs related to usage tracking, OpenTelemetry serialization, and tool invocation history.

Jul 29, 2026