NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI
PyPI, #1240 most downloaded on PyPI
Framework for evaluating stochastic code execution, especially code making use of LLMs
Latest stable
2.68.0
Last release 10 days ago
12 Nov 2026
Ships on a steady schedule
a new release about every 1 weeks
Nearly every release is documented
notes for 60 of the last 60 stable releases
1 version withdrawn
withdrawn after publishing
2 years old
310 releases · first in 2025
Drag across the ruler to compare, or tap a release to open it.
…parent_task_id fields, keeping the old names as deprecated aliases by @mpfaffenberger in #13074
OutputTokenLimitExceeded, and fall back to the next model in ModelFallback when a response is cut off at max_tokens by @DouweM in #12407RequestInput, put AskUser on the deferred pipeline, and mark deferred calls with kind and answered_by by @DouweM in #12757DelegationTasks.discover() and stop one damaged saved task from blocking DelegationTasks.opened() by @mpfaffenberger in #10140WorkspaceBackendSuite by @mpfaffenberger in #13070PromptInjectionDefender semantic classification as durable operations so it works on Temporal by @mpfaffenberger in #12424Event suffix and past-participle names, with agent_name and parent_task_id fields, keeping the old names as deprecated aliases by @mpfaffenberger in #13074pydantic_ai_harness, renaming num_results to max_results and max_duration_secs to max_duration_seconds by @mpfaffenberger in #13075CapabilityEvent fields to *_count, *_name, duration_ms and previous_item, keeping the old names as deprecated aliases by @mpfaffenberger in #13068CapabilityEvent outliers to <Subject><PastParticiple>Event and add STEP_PERSISTENCE_EVENTS by @mpfaffenberger in #13065RefineOutput harness capability by @mpfaffenberger in #13120ConversationStore with conversation_store by @DouweM in #12326GoogleModel and BedrockConverseModel streams by @ddanielcruzz in #13168SubAgents delegations to durable delegates as Temporal and DBOS child workflows by @mpfaffenberger in #12516RefineOutput harness capability by @mpfaffenberger in #13120cwd to Workspace.run by @mpfaffenberger in #12642WorkspaceBackendSuite by @mpfaffenberger in #13070DelegationTasks.discover() and stop one damaged saved task from blocking DelegationTasks.opened() by @mpfaffenberger in #10140WorkspaceBackendSuite reattach rule by @mpfaffenberger in #13081RequestInput, put AskUser on the deferred pipeline, and mark deferred calls with kind and answered_by by @DouweM in #12757OutputTokenLimitExceeded, and fall back to the next model in ModelFallback when a response is cut off at max_tokens by @DouweM in #12407CancellationToken fired from a workflow signal to cancel a Temporal durable run by @DouweM in #7210GoogleDecisionsModel backend for the Gemini Decisions API by @dsfaccini in #13181vendor/model ids through one case-insensitive family map on OpenAICompatibleProvider, LiteLLMProvider and TogetherProvider by @DouweM in #12743thinking to reasoning_effort for Z.AI GLM and DeepSeek V3.2 on Bedrock by @EdAbati in #11491Permissions remembered approvals inside durable execution by @mpfaffenberger in #12410AnthropicModel.count_tokens send the request's betas and raise its all-deferred tools UserError by @dsfaccini in #13098AnthropicStaleThinkingBlockWarning, since Claude Fable 5.2 and Mythos 5.2 bind thinking blocks on every account by @dsfaccini in #13155PromptInjectionDefender semantic classification as durable operations so it works on Temporal by @mpfaffenberger in #12424ModelResponse.canonical_system through the Vercel AI and AG-UI event streams by @DouweM in #13134max_tokens above a Claude model's maximum output to it instead of sending it by @DouweM in #12427TieredCompaction can't reach target_tokens because its tiers keep more than the target by @DouweM in #12650SummarizingCompaction(keep_user_messages=True) copies displace instead of dropping them by @DouweM in #12652genai-prices floor to 0.2.3 so Anthropic fast mode responses are priced at fast mode rates by @dsfaccini in #13104Full Changelog: v2.67.0...v2.68.0
Your coding agent can read these notes before it upgrades. Set up the MCP server →
Support SubAgents(dynamic_agents=...) under durable execution by @mpfaffenberger in #12418
SubAgents(dynamic_agents=...) under durable execution by @mpfaffenberger in #12418ArtifactScanner phrase matching and flag hidden variation selectors by @mpfaffenberger in #11475RepoContext, add walk_up, and have Coder build on them by @DouweM in #11691RepoContext(nested_max_chars=...) by @DouweM in #11699mimetypes knows, with the format taken from it by @arjun-krishna1 in #12365provider_settings by @adtyavrdhn in #7807framing_announcement setting and only announce the <system> voice tag once a conversation needs it by @DouweM in #12661RepoContext instruction files when the model writes or edits in a directory by @DouweM in #11701InputGuardrail, OutputGuardrail and ToolGuardrail verdicts as durable operations by @mpfaffenberger in #12428CapabilityCreation run when directory is below 'operator' authority by @DouweM in #12558history_digest through openai_previous_response_id by @DouweM in #12575CodeMode tools named after Python builtins, which the sandbox called instead of the tool by @adtyavrdhn in #12698ModelProfile.may_tokenize_special_tokens_in_content by @DouweM in #12556serialization_failure for durable capability results by @maksimzayats in #11057FileSearchTool.max_num_results to OpenAI Responses by @adtyavrdhn in #12728UserError when validation requests approval for an external tool, keep single-model tool arguments through durable re-validation, and document concurrent DBOS runs and shared MCP http_clients by @dsfaccini in #12958StepPersistence tool call that ends in ModelRetry or ToolFailed as failed instead of leaving it started by @dsfaccini in #12638UserError from StreamedRunResult.result when a run_stream() run stops while finishing by @dsfaccini in #13087CodeMode remote sandboxes over gRPC with pydantic-monty 1.5.0 by @dmontagu in #12673httpx with httpx2 in pydantic-ai-harness by @Kludex in #13026KnownModelName and the provider model-name literals to the models a user would pick by @dsfaccini in #12817dequeue and set_enqueue_priority to change a message queued with enqueue before delivery by @adtyavrdhn in #12345SubAgents(dynamic_agents=...) under durable execution by @mpfaffenberger in #12418child_workflow tool metadata in DBOSDurability by @mpfaffenberger in #12445TrajectoryJudge evaluations as a durable operation when the judge has an id by @mpfaffenberger in #12414pydantic-ai in their gRPC user agent by @tamirdresher in #12493client to ModalSandbox and ModalSandboxBackend by @adtyavrdhn in #12501ArtifactScanner phrase matching and flag hidden variation selectors by @mpfaffenberger in #11475client to E2BSandbox and E2BSandboxBackend by @adtyavrdhn in #12500Workspace.authority_of by @DouweM in #12473openai_include_unmapped_response_data to keep Chat Completions response data Pydantic AI doesn't map by @DouweM in #12034RepoContext, add walk_up, and have Coder build on them by @DouweM in #11691RepoContext(nested_max_chars=...) by @DouweM in #11699fast and ultrafast service tiers in openai_service_tier by @adtyavrdhn in #12527mimetypes knows, with the format taken from it by @arjun-krishna1 in #12365KeenableSearch capability for keyless web research by @ilya-bogin-keenable in #9079gateway/bedrock:global.moonshotai.kimi-k3 and gateway/bedrock:global.moonshotai.kimi-k3.1 to KnownModelName by @dsfaccini in #12869web_search_20260318 and web_fetch_20260318 via provider_settings by @adtyavrdhn in #12610provider_settings by @adtyavrdhn in #7807SubAgents(task_control=...) under Temporal and DBOS durability by @mpfaffenberger in #12521gpt-image-2.6-flare and gpt-image-2.6-sunburst image models by @dsfaccini in #12820claude-fable-5-2) and Claude Mythos 5.2 (claude-mythos-5-2) support by @dsfaccini in #12970global.anthropic.claude-fable-5-2) and send per-message effort to it on OpenRouter by @dsfaccini in #13037gemini-4-carbon) on GoogleModel by @dsfaccini in #13057framing_announcement setting and only announce the <system> voice tag once a conversation needs it by @DouweM in #12661RepoContext instruction files when the model writes or edits in a directory by @DouweM in #11701InputGuardrail, OutputGuardrail and ToolGuardrail verdicts as durable operations by @mpfaffenberger in #12428crun 1.14.1 in DockerSandbox and build AsyncDaytona off the event loop by @adtyavrdhn in #12509Advisor, SmartSearch and ConversationSearch tool I/O as durable operations by @mpfaffenberger in #12415FallbackModel attempts in the usage BrowserUse reports to browser-use by @mpfaffenberger in #11879BoxdSandbox fields in the agent spec schema by @adtyavrdhn in #12518DaytonaSandbox fields in the agent spec schema by @adtyavrdhn in #12517RenderSandbox fields in the agent spec schema by @adtyavrdhn in #12520SpritesSandbox fields in the agent spec schema by @adtyavrdhn in #12519$ref key as JSON text, which Gemini otherwise rejects by @DouweM in #12429retry_after before retrying the model request activity by @DouweM in #12106NativeOutput or PromptedOutput text that fails validation, instead of retrying by @DouweM in #12406ModelQuotaExceededError for the Gemini API's daily quota and disabled billing, and Moonshot's exhausted balance by @DouweM in #12422SummarizingCompaction, SlidingWindowCompaction and RelevanceCompaction by recording them again as a ToolAvailabilityDeltaPart by @DouweM in #11773on_complete on the current OpenTelemetry span by @DouweM in #12541There is no current event loop on Python 3.12 and 3.13 by @DouweM in #12540max_tokens setting to DeepSeek as max_tokens instead of max_completion_tokens by @dsfaccini in #12385WebSearchTool from non-compound GroqModel profiles so web search fallbacks run by @dsfaccini in #12386UserError from JsonSchemaTransformer on nodes that are not JSON Schemas by @adtyavrdhn in #8620Shell job's log within max_log_bytes when the job ends as its log watcher starts by @dsfaccini in #12578RepoContext.home_dir spelled through a symlink by @dsfaccini in #12614FunctionSignature type one name and one definition in rendered tool catalogs by @adtyavrdhn in #12513ClampOversizedMessages and ClearToolResults edits without the item ID that makes OpenAI Responses ignore them by @DouweM in #12561CapabilityCreation run when directory is below 'operator' authority by @DouweM in #12558history_digest through openai_previous_response_id by @DouweM in #12575CodeMode and IPythonCodeMode by @dmontagu in #12669CodeMode tools named after Python builtins, which the sandbox called instead of the tool by @adtyavrdhn in #12698max_retries raising a TypeError on AsyncAnthropicBedrock clients with anthropic 2.6 by @dsfaccini in #12707google-genai<3 so fresh installs keep GoogleModel on the v1beta Gemini API by @dsfaccini in #12700event_stream_handler the recorded events by @dsfaccini in #12675FileSystem search_files patterns whose repeat count overflows re, such as a{99999999999}, instead of aborting the run by @dsfaccini in #12726Shell job's supervisor and log watcher outliving a command that removes its job directory by @dsfaccini in #12730ModelProfile.may_tokenize_special_tokens_in_content by @DouweM in #12556CerebrasProvider reasoning flags for Qwen 3.8, Kimi K3 and GLM-5.3 by @DouweM in #12746CodeMode setup errors to callers, correct timeout advice, and reset local and remote sessions after memory/output limits by @dmontagu in #12689SpritesSandboxBackend.run raising a TypeError for a non-number timeout by @dsfaccini in #12729Shell job supervisor hang when a command removes its job directory or makes it read-only by @dsfaccini in #12814User-Agent header on Google requests by @dsfaccini in #12583Shell job whose command makes its job directory read-only, and stop it without a retry by @dsfaccini in #12830Shell failing to stop a macOS job whose process group holds only zombies, and keep a failed call's own error when its stop fails too by @dsfaccini in #12858gh-aw engine putting the checkout on sys.path ahead of installed packages by @dsfaccini in #12818serialization_failure for durable capability results by @maksimzayats in #11057FileSearchTool.max_num_results to OpenAI Responses by @adtyavrdhn in #12728extra_headers user-agent in any spelling in place of ours on OpenAI, Anthropic and Groq requests by @dsfaccini in #12871anthropic-beta header from extra_headers with the other Anthropic betas however its name is capitalized by @dsfaccini in #12875extra_headers as one header line on Google requests by @dsfaccini in #12877default_headers betas with ours, and accept a user-agent in batch extra_headers in any spelling by @dsfaccini in #12878UserError when validation requests approval for an external tool, keep single-model tool arguments through durable re-validation, and document concurrent DBOS runs and shared MCP http_clients by @dsfaccini in #12958thinking=False on more OpenRouter mandatory-reasoning routes by @DouweM in #12744$defs in gpt-oss tool schemas by @DouweM in #12759ToolOutput for gpt-oss on BedrockMantleProvider by not forcing tool_choice by @DouweM in #12761::ffff:0:0:0/96, 100:0:0:1::/64, 2001:2::/48, 3fff::/20 and 5f00::/16 as private addresses by @dsfaccini in #12953shell foreground wait replay the same workspace calls under durable execution by @mpfaffenberger in #12978task_backgrounded for a background task that settles before delegate_task resumes, as on durable replay by @dsfaccini in #13008tool_call_id on the tool call arguments the special-token restorer sends when a stream ends by @dsfaccini in #13062ModelAPIError.retry_after from Gemini's RetryInfo.retryDelay by Note truncated.
GHSA-m5vj-8mcw-82qx (high): web_fetch_tool and the WebFetch capability's local fallback re-indented a JSON response on the event loop before applying
web_fetch_tool and the WebFetch capability's local fallback re-indented a JSON response on the event loop before applying the returned-content limit, so a deeply nested document within the download size limit could expand to hundreds of times its size, using large amounts of memory and delaying other work in the same process. You are affected if an attacker can choose or influence the URLs the tool fetches. Provider-native web fetching is not affected. Fixed in v2.66.0 and v1.107.10 (pydantic-ai, pydantic-ai-slim).private_key pattern in Pydantic AI Harness's secret redaction backtracked exponentially on crafted text, so a few hundred characters that begin like a private key could occupy the event loop for minutes or longer. You are affected if you use redact_secrets or secret_data() with the private_key pattern, which is included by default, on text an attacker can influence. Fixed in pydantic-ai-harness 0.66.0.169.254.0.23, so someone who can supply a URL the application downloads could have it fetch the instance's temporary role credentials. You are affected if you run on Tencent Cloud and allow local URLs, through force_download='allow-local' on a FileUrl or web_fetch_tool(allow_local_urls=True). Fixed in v2.66.0 and v1.107.10 (pydantic-ai, pydantic-ai-slim).extra_headers and extra_body embedding settings, so credentials passed through them could be read by anyone with access to the exported traces. You are affected if you instrument embedding models and pass secrets through those settings. Fixed in v2.66.0 and v1.107.10 (pydantic-ai, pydantic-ai-slim). Restrict access to historical traces and rotate credentials they exposed.$defs and $ref, a small tool schema whose definitions reference one another repeatedly expanded exponentially when inlined on the event loop, which can take minutes of CPU time and gigabytes of memory. You are affected if you serve such a model through the AG-UI adapter, where clients submit their own frontend tool schemas. Fixed in v2.66.0 and v1.107.10 (pydantic-ai, pydantic-ai-slim).DecisionModel subclasses such as OpenAIDecisionsModel, and TypeSafeModel) resolved every $ref in a tool's argument schema again at each place it appears, so a small crafted schema could take exponential time and memory on the event loop. You are affected if you use a decision model from v2.46.0 onward with tools whose schemas an attacker can supply, most directly through an AG-UI endpoint. Fixed in v2.66.0 (pydantic-ai, pydantic-ai-slim).VercelAIAdapter or AGUIAdapter. Fixed in v2.66.0 and v1.107.10 (pydantic-ai, pydantic-ai-slim).s3:// object or uploaded file with the application's credentials. You are affected if tool results can contain data third parties control, that history is stored and reloaded, held on the server by a UI adapter, or carried by Temporal, DBOS or Prefect durable execution, and your model or its provider fetches or resolves such files; the advisory lists which do. Fixed in v2.66.0 and v1.107.10 (pydantic-ai, pydantic-ai-slim).allowed_domains on the local web-fetch tool, and PlaywrightBrowser egress policies whose allowlist bounds every request kind, looked a refused hostname up in DNS before rejecting it, so data encoded in the hostname could reach that domain's nameserver. Domain lists are now checked before resolution. You are affected only if you rely on such an allowlist while the model processes content untrusted parties can influence. Fixed in v2.66.0 and v1.107.10 (pydantic-ai, pydantic-ai-slim), and pydantic-ai-harness 0.66.0.MCP capability configured with a server URL and no explicit id derived its identifier from the URL's hostname and last path segment, and the AG-UI and Vercel AI adapters sent that identifier to the frontend with MCP Apps tool results, so a credential in that segment could reach users of the chat frontend. This affects v2.57.0 up to but not including v2.60.0, and was fixed in v2.60.0; it is disclosed with this batch. Rotate MCP server credentials that were carried in the URL path.20260209 web tools directly on the Anthropic API when citations are requested by @dsfaccini in #12278FunctionToolset(requires_approval=True) to Tool instances passed to the toolset by @dsfaccini in #12378SummarizingCompaction(keep_user_messages=True) copies only by the marker on their text by @DouweM in #12316OpenSandbox for self-hosted agent workspaces by @adtyavrdhn in #11841pydantic_ai_harness by @DouweM in #11332Pixeltable tools and memory storage to Harness by @pierrebrunelle in #9369RenderSandbox to run agent workspaces in Render sandboxes by @ojusave in #10238include_instructions to the MCP capability so server instructions can be left out of the agent's instructions by @DouweM in #12269segment timestamps with output_speech_timestamps by @DouweM in #11782retries field to SummarizingCompaction by @gouyu-hou in #12315Workspace.authority and AuthorityWorkspace for workspaces whose files the operator does not control by @adtyavrdhn in #11836ModalSandbox failing with AttributeError on Modal 1.6, which removed FilesystemExecutionError by @adtyavrdhn in #1224720260209 web tools directly on the Anthropic API when citations are requested by @dsfaccini in #12278CohereModel by @Julien-ser in #6100tool_search and load_capability message history by @pydanty in #7933$defs in tool schemas for GLM-5.3 on Together, Nebius, Fireworks, Vercel, Ollama, Hugging Face, Cerebras, Mistral and OpenAI-compatible gateways by @pydanty in #12138RepoContext from following symlinks out of the working directory or to files that aren't instruction files by @DouweM in #12044manage_message_history='server' by @DouweM in #12313FunctionToolset(requires_approval=True) to Tool instances passed to the toolset by @dsfaccini in #12378SummarizingCompaction(keep_user_messages=True) copies only by the marker on their text by @DouweM in #12316SummarizingCompaction, SlidingWindowCompaction and RelevanceCompaction, recording every removed load again as a CapabilityLoadDeltaPart by @DouweM in #12270openai_chat_send_back_thinking_parts='field' mode by @DouweM in #12309RepairToolArguments repairs back to the ToolCallPart in history by @DouweM in #12319AnthropicModel.compact_messages()'s focus prompt by @DouweM in #11915SqliteStepStore and SqliteMediaStore connections whose WAL switch fails on a locked database by @DouweM in #12147embeddings attribute when include_content=True by @pydanty in #8295OpenAIChatModel's batch hint, OpenAI realtime models and harness Advisor on provider names by @DouweM in #11796Full Changelog: v2.65.0...v2.66.0
Protect the gh-aw canonical session capture from agent tools by @dsfaccini in #10606
gh-aw canonical session capture from agent tools by @dsfaccini in #10606AskUserAnswer by @dsfaccini in #11664BrowserUse model calls toward the host run's usage and UsageLimits by @mpfaffenberger in #11860OpenAIModel or MCPServerStdio by @DouweM in #12035ModelFallback when a model refuses, and let ContentFilterError in fallback_on match refusals by @DouweM in #12039ModelQuotaExceededError instead of ModelRateLimitError for Z.ai balance and plan-limit 429s by @DouweM in #12042'12h' prompt cache retention on the OpenAI Responses API for GPT-5.6 and later by @DouweM in #12046non_retryable_model_errors by @DouweM in #12107ConversationStore, which keeps a conversation as a log of messages stored once, and Conversation.revision to catch conflicting saves by @DouweM in #10658AskUserAnswer by @dsfaccini in #11664anthropic_web_tools_allowed_callers to set allowed_callers on Anthropic's web tools by @aweis89 in #10489'12h' prompt cache retention on the OpenAI Responses API for GPT-5.6 and later by @DouweM in #12046OpenAIModel or MCPServerStdio by @DouweM in #12035ModelFallback when a model refuses, and let ContentFilterError in fallback_on match refusals by @DouweM in #12039ModelContextWindowExceededError and ModelQuotaExceededError for Bedrock errors on AsyncAnthropicBedrock by @DouweM in #12040ContentFilterError instead of TypeError when Azure filters a streamed OpenAIChatModel prompt by @DouweM in #12041RepoContext instruction files after compaction, and stop repeating notes still in history by @DouweM in #11704ContentFilterError when the provider also reports a finish reason by @DouweM in #12036ModelQuotaExceededError instead of ModelRateLimitError for Z.ai balance and plan-limit 429s by @DouweM in #12042gh-aw canonical session capture from agent tools by @dsfaccini in #10606PlaywrightBrowser's Chromium and driver when an anyio scope cancels a run by @dsfaccini in #12047Coder to match the surrounding code's style instead of applying SOLID by @mpfaffenberger in #12120Coder tool's error to the model once its retries run out instead of ending the run by @mpfaffenberger in #12121*args and **kwargs docstring descriptions on tool parameters by @Aryapatel-007 in #12126Keyword Args and Other Parameters docstring sections by @roydonsequeira in #10035aws_lambda agent loop finish its own cleanup at the grace deadline by @dsfaccini in #12140auth='' as unset alongside client in GitHub, Notion, Ordinal, PostHog, Pylon, and Slack by @mpfaffenberger in #11852BrowserUse model calls toward the host run's usage and UsageLimits by @mpfaffenberger in #11860FileSystem tools emit their events under PrefixTools by @mpfaffenberger in #11867cache='1h' to Claude Haiku 5.5 on Bedrock instead of snapping it to '5m' by @DouweM in #12045<context> elements in user and tool content as not from the application by @DouweM in #12050ContentFilterError or a UserError subclass by @DouweM in #12105non_retryable_model_errors by @DouweM in #12107close() workaround and require daytona>=0.226.1 by @mislavivanda in #11977Full Changelog: v2.64.0...v2.65.0
<!-- Release notes generated using configuration in .github/release.yml at a56a6d6f138d1e3a9c45a9697ea513bef47a6cea -->
gh-aw canonical session capture from agent tools by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/10606AskUserAnswer by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11664BrowserUse model calls toward the host run's usage and UsageLimits by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11860OpenAIModel or MCPServerStdio by @DouweM in https://github.com/pydantic/pydantic-ai/pull/12035ModelFallback when a model refuses, and let ContentFilterError in fallback_on match refusals by @DouweM in https://github.com/pydantic/pydantic-ai/pull/12039ModelQuotaExceededError instead of ModelRateLimitError for Z.ai balance and plan-limit 429s by @DouweM in https://github.com/pydantic/pydantic-ai/pull/12042'12h' prompt cache retention on the OpenAI Responses API for GPT-5.6 and later by @DouweM in https://github.com/pydantic/pydantic-ai/pull/12046non_retryable_model_errors by @DouweM in https://github.com/pydantic/pydantic-ai/pull/12107ConversationStore, which keeps a conversation as a log of messages stored once, and Conversation.revision to catch conflicting saves by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10658AskUserAnswer by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11664anthropic_web_tools_allowed_callers to set allowed_callers on Anthropic's web tools by @aweis89 in https://github.com/pydantic/pydantic-ai/pull/10489'12h' prompt cache retention on the OpenAI Responses API for GPT-5.6 and later by @DouweM in https://github.com/pydantic/pydantic-ai/pull/12046OpenAIModel or MCPServerStdio by @DouweM in https://github.com/pydantic/pydantic-ai/pull/12035ModelFallback when a model refuses, and let ContentFilterError in fallback_on match refusals by @DouweM in https://github.com/pydantic/pydantic-ai/pull/12039ModelContextWindowExceededError and ModelQuotaExceededError for Bedrock errors on AsyncAnthropicBedrock by @DouweM in https://github.com/pydantic/pydantic-ai/pull/12040ContentFilterError instead of TypeError when Azure filters a streamed OpenAIChatModel prompt by @DouweM in https://github.com/pydantic/pydantic-ai/pull/12041RepoContext instruction files after compaction, and stop repeating notes still in history by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11704ContentFilterError when the provider also reports a finish reason by @DouweM in https://github.com/pydantic/pydantic-ai/pull/12036ModelQuotaExceededError instead of ModelRateLimitError for Z.ai balance and plan-limit 429s by @DouweM in https://github.com/pydantic/pydantic-ai/pull/12042gh-aw canonical session capture from agent tools by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/10606PlaywrightBrowser's Chromium and driver when an anyio scope cancels a run by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/12047Coder to match the surrounding code's style instead of applying SOLID by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/12120Coder tool's error to the model once its retries run out instead of ending the run by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/12121*args and **kwargs docstring descriptions on tool parameters by @Aryapatel-007 in https://github.com/pydantic/pydantic-ai/pull/12126Keyword Args and Other Parameters docstring sections by @roydonsequeira in https://github.com/pydantic/pydantic-ai/pull/10035aws_lambda agent loop finish its own cleanup at the grace deadline by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/12140auth='' as unset alongside client in GitHub, Notion, Ordinal, PostHog, Pylon, and Slack by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11852BrowserUse model calls toward the host run's usage and UsageLimits by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11860FileSystem tools emit their events under PrefixTools by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11867cache='1h' to Claude Haiku 5.5 on Bedrock instead of snapping it to '5m' by @DouweM in https://github.com/pydantic/pydantic-ai/pull/12045<context> elements in user and tool content as not from the application by @DouweM in https://github.com/pydantic/pydantic-ai/pull/12050ContentFilterError or a UserError subclass by @DouweM in https://github.com/pydantic/pydantic-ai/pull/12105non_retryable_model_errors by @DouweM in https://github.com/pydantic/pydantic-ai/pull/12107close() workaround and require daytona>=0.226.1 by @mislavivanda in https://github.com/pydantic/pydantic-ai/pull/11977Full Changelog: https://github.com/pydantic/pydantic-ai/compare/v2.64.0...v2.65.0
<!-- Release notes generated using configuration in .github/release.yml at cde1d10f88ce6d0ab4743e226d51a9ac65810f50 -->
<!-- Release notes generated using configuration in .github/release.yml at cde1d10f88ce6d0ab4743e226d51a9ac65810f50 -->
ModelHTTPError, so fallback moves on by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11934Model.accepts_server_state by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11774json-repair an optional dependency, needed only by Coder and RepairToolArguments by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11768Coder's tools by default, with Coder(strict=True) to opt in by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11743RepoContext instruction file between a traversed path and the working directory, without repeating the root file by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11687output_speech_timestamps to report word timing of realtime speech, on Azure AI Voice Live by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11685ModelFallback capability constructible from agent specs by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11679first_event_timeout model setting so a stream that never starts can fall back by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11582OnlineEvaluation evaluators once per durable run instead of on every replay by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11577Planning plan in the plan tools' results instead of a per-request reminder copy by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11521Memory and Skills with ArtifactScanner by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11470AnthropicModel.count_tokens rejecting all-deferred tool requests and undercounting server tools on the Anthropic API by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11343pydantic_evals by @dmontagu in https://github.com/pydantic/pydantic-ai/pull/10292RealtimeEvent from pydantic_ai.agent.__all__ by @adtyavrdhn in https://github.com/pydantic/pydantic-ai/pull/10189RunContext.load_capability to load on-demand capabilities from code by @adtyavrdhn in https://github.com/pydantic/pydantic-ai/pull/10090ModelFallback capability, a per-attempt prepare_model_request hook, and RetryModelRequest by @DouweM in https://github.com/pydantic/pydantic-ai/pull/9348@agent.system_prompt(dynamic=True) append its changes with on_change='append' to keep the cached prompt by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11445DetectLoops tool results with the response that made the calls by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11551BoxdSandbox: a workspace that runs Coder, Shell, FileSystem and every other harness capability in a boxd machine by @MichielMAnalytics in https://github.com/pydantic/pydantic-ai/pull/11367ModelFallback capability, a per-attempt prepare_model_request hook, and RetryModelRequest by @DouweM in https://github.com/pydantic/pydantic-ai/pull/9348Reminder(scope='conversation') for reminders that stay in the message history by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11642output_speech_timestamps to report word timing of realtime speech, on Azure AI Voice Live by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11685ModelRequestContext durable capability operations across agent construction by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11776ModelFallback capability constructible from agent specs by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11679output_speech_timestamps by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11784vcs.* and cicd.* attributes by @dmontagu in https://github.com/pydantic/pydantic-ai/pull/10294pydantic_evals by @dmontagu in https://github.com/pydantic/pydantic-ai/pull/10292logfire.scorer.score_type and logfire.scorer.run_id to online evaluation result events by @dmontagu in https://github.com/pydantic/pydantic-ai/pull/10293test.suite.name and test.case.name attributes to Pydantic Evals spans by @dmontagu in https://github.com/pydantic/pydantic-ai/pull/11538Coder's tools by default, with Coder(strict=True) to opt in by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11743json-repair an optional dependency, needed only by Coder and RepairToolArguments by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11768Coder to read what the task names, check stated requirements, and delegate only when asked by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11811AskUser answerers receive RunContext by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11662Memory and Skills with ArtifactScanner by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11470ModelResponse.model_id recording the model the agent sent each request to by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11807Dataset files under the function's name by @dmontagu in https://github.com/pydantic/pydantic-ai/pull/10425RepoContext instructions when nested_traversal is on by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11698first_event_timeout model setting so a stream that never starts can fall back by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11582sticky to ModelFallback to keep a conversation on the fallback model that answered by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11330RunContext.load_capability to load on-demand capabilities from code by @adtyavrdhn in https://github.com/pydantic/pydantic-ai/pull/10090SpanTree attributes during JSON round-trips and mark omitted Google embedding and ElevenLabs usage as unmeasured to avoid false zero automatic costs by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11912botocore predates outputConfig by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11694IPythonKernel client's zmq context on shutdown with jupyter-client before 8.7 by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11716CodeMode Tool Search corpus with search_tools, so a native deferred tool no longer declares the sandboxed ones by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11638Planning plan in the plan tools' results instead of a per-request reminder copy by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11521OnlineEvaluation evaluators once per durable run instead of on every replay by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11577AnthropicModel.count_tokens rejecting all-deferred tool requests and undercounting server tools on the Anthropic API by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11343CodeExecutionTool on the Bedrock AnthropicModel clients by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11790cache setting for GPT-5.6 and later with the Azure OpenAI v1 API by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11749Shell's job files in a private temp directory when the workspace isn't writable by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11809$refs under propertyNames and contains for OpenRouter Google models and custom prefer_inlined_defs transformers by @jukebox in https://github.com/pydantic/pydantic-ai/pull/10859BedrockJsonSchemaTransformer by @adtyavrdhn in https://github.com/pydantic/pydantic-ai/pull/11831RealtimeEvent from pydantic_ai.agent.__all__ by @adtyavrdhn in https://github.com/pydantic/pydantic-ai/pull/10189BackgroundTools results in the history when a run pauses for deferred tools by @wangrat in https://github.com/pydantic/pydantic-ai/pull/11752FileSystem edit_file edit non-UTF-8 text files that read_file reads, honoring expected_hash by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11851SubAgents children that share usage run under the parent's usage_limits by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11853AgentSpec schema fields for Advisor, SummarizingCompaction, ManagedPrompt and PyaiDocs by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11857asyncpg pool type-check as PostgresMemoryStore's PostgresPool by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11856AbstractCapability base fields id, description and defer_loading by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11858AgentSpec schema fields for CodeMode, ModalSandbox, and Skills by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11859SubAgents under Temporal for a named agent with its own model and self with a model menu by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11862prefix_tools() by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11863result.usage() warn instead of raising TypeError by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11884content from OpenAI-compatible Chat Completions servers by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11885BedrockConverseModel call or stream read when its request is cancelled or times out by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11850SystemPromptParts visible: false in OTel spans of requests the model no longer sees them on by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11772Model.accepts_server_state by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11774system entry on Claude Haiku 5.5, so it doesn't read the entry as tool output by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11785ModelResponse.canonical_system through Vercel AI and AG-UI message round-trips by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11794FileUnderstanding, and let prepare_model_request persist history changes by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11802RepoContext instruction files above the working directory as ../AGENTS.md instead of by absolute path by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11689on_run_error, shielded capability cleanup and the model stream close run when an anyio scope cancels a run by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/10479DynamicWorkflow sub-agents under the parent run's usage_limits instead of the 50-request default by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11962web_fetch leaking script/style source into markdown content by @pydanty[bot] in https://github.com/pydantic/pydantic-ai/pull/8329StepPersistence failure records when an anyio scope cancels a run by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11993max_retries in OpenAI, Anthropic and Gemini batch requests instead of raising UserError by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11931ModelHTTPError, so fallback moves on by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11934RepoContext instruction file between a traversed path and the working directory, without repeating the root file by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11687Full Changelog: https://github.com/pydantic/pydantic-ai/compare/v2.63.0...v2.64.0
ModelHTTPError, so fallback moves on by @DouweM in #11934Model.accepts_server_state by @DouweM in #11774json-repair an optional dependency, needed only by Coder and RepairToolArguments by @mpfaffenberger in #11768Coder's tools by default, with Coder(strict=True) to opt in by @mpfaffenberger in #11743RepoContext instruction file between a traversed path and the working directory, without repeating the root file by @DouweM in #11687output_speech_timestamps to report word timing of realtime speech, on Azure AI Voice Live by @DouweM in #11685ModelFallback capability constructible from agent specs by @DouweM in #11679first_event_timeout model setting so a stream that never starts can fall back by @DouweM in #11582OnlineEvaluation evaluators once per durable run instead of on every replay by @DouweM in #11577Planning plan in the plan tools' results instead of a per-request reminder copy by @DouweM in #11521Memory and Skills with ArtifactScanner by @mpfaffenberger in #11470AnthropicModel.count_tokens rejecting all-deferred tool requests and undercounting server tools on the Anthropic API by @dsfaccini in #11343pydantic_evals by @dmontagu in #10292RealtimeEvent from pydantic_ai.agent.__all__ by @adtyavrdhn in #10189RunContext.load_capability to load on-demand capabilities from code by @adtyavrdhn in #10090ModelFallback capability, a per-attempt prepare_model_request hook, and RetryModelRequest by @DouweM in #9348@agent.system_prompt(dynamic=True) append its changes with on_change='append' to keep the cached prompt by @DouweM in #11445DetectLoops tool results with the response that made the calls by @mpfaffenberger in #11551BoxdSandbox: a workspace that runs Coder, Shell, FileSystem and every other harness capability in a boxd machine by @MichielMAnalytics in #11367ModelFallback capability, a per-attempt prepare_model_request hook, and RetryModelRequest by @DouweM in #9348Reminder(scope='conversation') for reminders that stay in the message history by @DouweM in #11642output_speech_timestamps to report word timing of realtime speech, on Azure AI Voice Live by @DouweM in #11685ModelRequestContext durable capability operations across agent construction by @DouweM in #11776ModelFallback capability constructible from agent specs by @DouweM in #11679output_speech_timestamps by @DouweM in #11784vcs.* and cicd.* attributes by @dmontagu in #10294pydantic_evals by @dmontagu in #10292logfire.scorer.score_type and logfire.scorer.run_id to online evaluation result events by @dmontagu in #10293test.suite.name and test.case.name attributes to Pydantic Evals spans by @dmontagu in #11538Coder's tools by default, with Coder(strict=True) to opt in by @mpfaffenberger in #11743json-repair an optional dependency, needed only by Coder and RepairToolArguments by @mpfaffenberger in #11768Coder to read what the task names, check stated requirements, and delegate only when asked by @mpfaffenberger in #11811AskUser answerers receive RunContext by @dsfaccini in #11662Memory and Skills with ArtifactScanner by @mpfaffenberger in #11470ModelResponse.model_id recording the model the agent sent each request to by @DouweM in #11807Dataset files under the function's name by @dmontagu in #10425RepoContext instructions when nested_traversal is on by @DouweM in #11698first_event_timeout model setting so a stream that never starts can fall back by @DouweM in #11582sticky to ModelFallback to keep a conversation on the fallback model that answered by @DouweM in #11330RunContext.load_capability to load on-demand capabilities from code by @adtyavrdhn in #10090SpanTree attributes during JSON round-trips and mark omitted Google embedding and ElevenLabs usage as unmeasured to avoid false zero automatic costs by @dsfaccini in #11912botocore predates outputConfig by @DouweM in #11694IPythonKernel client's zmq context on shutdown with jupyter-client before 8.7 by @DouweM in #11716CodeMode Tool Search corpus with search_tools, so a native deferred tool no longer declares the sandboxed ones by @DouweM in #11638Planning plan in the plan tools' results instead of a per-request reminder copy by @DouweM in #11521OnlineEvaluation evaluators once per durable run instead of on every replay by @DouweM in #11577AnthropicModel.count_tokens rejecting all-deferred tool requests and undercounting server tools on the Anthropic API by @dsfaccini in #11343CodeExecutionTool on the Bedrock AnthropicModel clients by @DouweM in #11790cache setting for GPT-5.6 and later with the Azure OpenAI v1 API by @DouweM in #11749Shell's job files in a private temp directory when the workspace isn't writable by @mpfaffenberger in #11809$refs under propertyNames and contains for OpenRouter Google models and custom prefer_inlined_defs transformers by @jukebox in #10859BedrockJsonSchemaTransformer by @adtyavrdhn in #11831RealtimeEvent from pydantic_ai.agent.__all__ by @adtyavrdhn in #10189BackgroundTools results in the history when a run pauses for deferred tools by @wangrat in #11752FileSystem edit_file edit non-UTF-8 text files that read_file reads, honoring expected_hash by @mpfaffenberger in #11851SubAgents children that share usage run under the parent's usage_limits by @mpfaffenberger in #11853AgentSpec schema fields for Advisor, SummarizingCompaction, ManagedPrompt and PyaiDocs by @mpfaffenberger in #11857asyncpg pool type-check as PostgresMemoryStore's PostgresPool by @mpfaffenberger in #11856AbstractCapability base fields id, description and defer_loading by @mpfaffenberger in #11858AgentSpec schema fields for CodeMode, ModalSandbox, and Skills by @mpfaffenberger in #11859SubAgents under Temporal for a named agent with its own model and self with a model menu by @mpfaffenberger in #11862prefix_tools() by @mpfaffenberger in #11863result.usage() warn instead of raising TypeError by @DouweM in #11884content from OpenAI-compatible Chat Completions servers by @DouweM in #11885BedrockConverseModel call or stream read when its request is cancelled or times out by @DouweM in #11850SystemPromptParts visible: false in OTel spans of requests the model no longer sees them on by @DouweM in #11772Model.accepts_server_state by @DouweM in #11774system entry on Claude Haiku 5.5, so it doesn't read the entry as tool output by @DouweM in #11785ModelResponse.canonical_system through Vercel AI and AG-UI message round-trips by @DouweM in #11794FileUnderstanding, and let prepare_model_request persist history changes by @DouweM in #11802RepoContext instruction files above the working directory as ../AGENTS.md instead of by absolute path by @DouweM in #11689on_run_error, shielded capability cleanup and the model stream close run when an anyio scope cancels a run by @dsfaccini in #10479DynamicWorkflow sub-agents under the parent run's usage_limits instead of the 50-request default by @mpfaffenberger in #11962web_fetch leaking script/style source into markdown content by @pydanty[bot] in #8329StepPersistence failure records when an anyio scope cancels a run by @dsfaccini in #11993max_retries in OpenAI, Anthropic and Gemini batch requests instead of raising UserError by @DouweM in #11931ModelHTTPError, so fallback moves on by @DouweM in #11934RepoContext instruction file between a traversed path and the working directory, without repeating the root file by @DouweM in #11687Full Changelog: v2.63.0...v2.64.0
<!-- Release notes generated using configuration in .github/release.yml at e2fbcf609852600e2f2e2eabb18c0b233c49a937 -->
<!-- Release notes generated using configuration in .github/release.yml at e2fbcf609852600e2f2e2eabb18c0b233c49a937 -->
CodeMode to dynamic_catalog=True, so a revealed tool is appended instead of rewriting the cached prompt by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11358GoogleRealtimeConnection's dial as a keyword-only replay argument by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11515LLMReminder a default interval=10 instead of generating on every request by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11641canonical_system by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11504CombinedCapability load as one capability, with members keeping their global ordering by @adtyavrdhn in https://github.com/pydantic/pydantic-ai/pull/10321<system> tag means and escape it in user and tool text on models without mid-conversation system messages by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11506task_timeout to Dataset.evaluate to bound each task attempt by @dmontagu in https://github.com/pydantic/pydantic-ai/pull/10291ag-ui-protocol 1.0 by @adtyavrdhn in https://github.com/pydantic/pydantic-ai/pull/8627CodeMode to dynamic_catalog=True, so a revealed tool is appended instead of rewriting the cached prompt by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11358GoogleRealtimeConnection's dial as a keyword-only replay argument by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11515gpt-6.1-terra model support by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11393IPythonCodeMode) alongside the Monty-based CodeMode by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11570LLMReminder a default interval=10 instead of generating on every request by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11641DetectLoops capability to interrupt repeated-action loops by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11471manage_message_history to UI adapters so a server-held history isn't sent to the model twice by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11431pydantic_ai.framing module with wrap, neutralize_tags and register_tag, so extensions can frame their own tags by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10957canonical_system by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11504PostgresSpendStore so harness SpendLimits budgets can be shared through PostgreSQL by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11527CombinedCapability load as one capability, with members keeping their global ordering by @adtyavrdhn in https://github.com/pydantic/pydantic-ai/pull/10321CodeMode run_code description by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11459mistralai>=3.1.0 and drop the legacy httpx import so MistralModel imports on fresh installs by @adtyavrdhn in https://github.com/pydantic/pydantic-ai/pull/11549too_many_requests Responses failures on openai>=3.32.0 by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11532ToolOutputLimits spill writes in a durable_operation so replays do not write them again by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11460before_model_request hook sent when it rewrote the history with a copy of the request by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11501before_model_request hook loaded it by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11516BackgroundTools, RepoContext, TrajectoryJudge and Planning frames by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11530thinking change after an unanswered ModelSettingsDeltaPart instead of replacing it, so UI streams and hook-rebuilt requests keep it by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11514gemini-robotics-er realtime models see the image sent just before them by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11484PendingModelResponse by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11517AsyncAnthropicBedrockMantle on Claude Opus 5.5 by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11486AsyncAnthropicBedrockMantle by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11487AdvisorTool on AsyncAnthropicBedrockMantle, which rejects it by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11488CodeMode REPL state when repl_store load fails, instead of saving a fresh REPL over it by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11483RuntimeError: coroutine ignored GeneratorExit and an escaping KeyboardInterrupt on Agent.iter() teardown by @pydanty in https://github.com/pydantic/pydantic-ai/pull/10258wrap_run or on_run_error recovery result from run_stream() instead of raising AgentRunError by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11571<system> tag means and escape it in user and tool text on models without mid-conversation system messages by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11506SummarizingCompaction summarizer transcript by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11583allowed_patterns on the real path when search_files targets a single symlinked file by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11639None type arguments in Dataset by @pydanty in https://github.com/pydantic/pydantic-ai/pull/11254WebSearchTool and WebFetchTool on AsyncAnthropicBedrockMantle, which rejects them by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11489SummarizingCompaction summary that would make the history larger by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11649stream_transcripts(delta=True) when a transcript is corrected to empty by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11667OnlineEvaluation capability on an agent, not just the outermost by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11663model_settings when include_content=False by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11668AsyncAnthropicBedrockMantle the Bedrock Messages API, not Claude Platform on AWS, in comments and docstrings by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11482SpeechPart.id by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11675Full Changelog: https://github.com/pydantic/pydantic-ai/compare/v2.62.0...v2.63.0
CodeMode to dynamic_catalog=True, so a revealed tool is appended instead of rewriting the cached prompt by @DouweM in #11358GoogleRealtimeConnection's dial as a keyword-only replay argument by @DouweM in #11515LLMReminder a default interval=10 instead of generating on every request by @DouweM in #11641canonical_system by @DouweM in #11504CombinedCapability load as one capability, with members keeping their global ordering by @adtyavrdhn in #10321<system> tag means and escape it in user and tool text on models without mid-conversation system messages by @DouweM in #11506task_timeout to Dataset.evaluate to bound each task attempt by @dmontagu in #10291ag-ui-protocol 1.0 by @adtyavrdhn in #8627CodeMode to dynamic_catalog=True, so a revealed tool is appended instead of rewriting the cached prompt by @DouweM in #11358GoogleRealtimeConnection's dial as a keyword-only replay argument by @DouweM in #11515gpt-6.1-terra model support by @dsfaccini in #11393IPythonCodeMode) alongside the Monty-based CodeMode by @mpfaffenberger in #11570LLMReminder a default interval=10 instead of generating on every request by @DouweM in #11641DetectLoops capability to interrupt repeated-action loops by @mpfaffenberger in #11471manage_message_history to UI adapters so a server-held history isn't sent to the model twice by @DouweM in #11431pydantic_ai.framing module with wrap, neutralize_tags and register_tag, so extensions can frame their own tags by @DouweM in #10957canonical_system by @DouweM in #11504PostgresSpendStore so harness SpendLimits budgets can be shared through PostgreSQL by @DouweM in #11527CombinedCapability load as one capability, with members keeping their global ordering by @adtyavrdhn in #10321CodeMode run_code description by @mpfaffenberger in #11459mistralai>=3.1.0 and drop the legacy httpx import so MistralModel imports on fresh installs by @adtyavrdhn in #11549too_many_requests Responses failures on openai>=3.32.0 by @dsfaccini in #11532ToolOutputLimits spill writes in a durable_operation so replays do not write them again by @mpfaffenberger in #11460before_model_request hook sent when it rewrote the history with a copy of the request by @DouweM in #11501before_model_request hook loaded it by @DouweM in #11516BackgroundTools, RepoContext, TrajectoryJudge and Planning frames by @DouweM in #11530thinking change after an unanswered ModelSettingsDeltaPart instead of replacing it, so UI streams and hook-rebuilt requests keep it by @DouweM in #11514gemini-robotics-er realtime models see the image sent just before them by @DouweM in #11484PendingModelResponse by @DouweM in #11517AsyncAnthropicBedrockMantle on Claude Opus 5.5 by @DouweM in #11486AsyncAnthropicBedrockMantle by @DouweM in #11487AdvisorTool on AsyncAnthropicBedrockMantle, which rejects it by @DouweM in #11488CodeMode REPL state when repl_store load fails, instead of saving a fresh REPL over it by @mpfaffenberger in #11483RuntimeError: coroutine ignored GeneratorExit and an escaping KeyboardInterrupt on Agent.iter() teardown by @pydanty in #10258wrap_run or on_run_error recovery result from run_stream() instead of raising AgentRunError by @DouweM in #11571<system> tag means and escape it in user and tool text on models without mid-conversation system messages by @DouweM in #11506SummarizingCompaction summarizer transcript by @DouweM in #11583allowed_patterns on the real path when search_files targets a single symlinked file by @mpfaffenberger in #11639None type arguments in Dataset by @pydanty in #11254WebSearchTool and WebFetchTool on AsyncAnthropicBedrockMantle, which rejects them by @DouweM in #11489SummarizingCompaction summary that would make the history larger by @DouweM in #11649stream_transcripts(delta=True) when a transcript is corrected to empty by @DouweM in #11667OnlineEvaluation capability on an agent, not just the outermost by @DouweM in #11663model_settings when include_content=False by @DouweM in #11668AsyncAnthropicBedrockMantle the Bedrock Messages API, not Claude Platform on AWS, in comments and docstrings by @DouweM in #11482SpeechPart.id by @DouweM in #11675Full Changelog: v2.62.0...v2.63.0
Build Harness SystemReminders on request-scoped system prompts and deprecate cache_ttl by @DouweM in https://github.com/pydantic/pydantic-ai/pull/9712
<!-- Release notes generated using configuration in .github/release.yml at 79685d18af7c8ea174029f23f40eec7c4af18e2b -->
configuration_update and Anthropic per-message output_config, recorded as ModelSettingsDeltaPart by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11111ModelRequestRejectedError, ModelAuthenticationError and ModelNotFoundError model error categories by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11081before_model_request by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11103tool_prefix by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10376SummarizingCompaction summary as a CompactionPart instead of a system prompt, and send provider-less CompactionParts to every model by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10573SystemPromptParts on each provider's operator channel with session.send() by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11379on_change='append', and give an id-less URL MCPToolset its hostname as id by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10791Memory snapshots in history once per changed run, so requests stay append-only by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10903ctx.emit() by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11157SpeechPart instead of splitting it at each delegated tool call by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11187ctx.emit() events into the realtime session event stream by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11369ModelSettings: forced tool_choice raises on Gemini Live, parallel_tool_calls=False warns by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11066configuration_update system message by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11426GraphBuilder broadcast branches in .to() order instead of reversed, which can break replay of durable workflows paused across the upgrade by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11408on_change='append' instruction blocks behind a second cache point after the static instructions by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11351SystemReminders on request-scoped system prompts and deprecate cache_ttl by @DouweM in https://github.com/pydantic/pydantic-ai/pull/9712CodeMode(monty_sandbox=...), which takes a remote Monty URL or a function that builds the pool, and deprecate monty_sandbox_url by @adtyavrdhn in https://github.com/pydantic/pydantic-ai/pull/11141configuration_update and Anthropic per-message output_config, recorded as ModelSettingsDeltaPart by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11111ModelRequestRejectedError, ModelAuthenticationError and ModelNotFoundError model error categories by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11081conversation_id_from_run_id from pydantic_ai.durable_exec for third-party durability engines by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11338child_workflow tool metadata in TemporalDurability by @YHallouard in https://github.com/pydantic/pydantic-ai/pull/7329tool_choice subsets with allowed_tools on OpenAI Chat Completions to preserve the prompt cache by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10396before_model_request by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11103tool_prefix by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10376SummarizingCompaction summary as a CompactionPart instead of a system prompt, and send provider-less CompactionParts to every model by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10573tool_prefix to the MCP capability by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11364SubAgents run delegations as tasks the model can background, steer, answer and cancel, behind task_control by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10893Coder(task_control=) to turn on SubAgents task control by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11007StructuredContent user content, sent as its value to decision models by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10529RequestUsage.cost by @adtyavrdhn in https://github.com/pydantic/pydantic-ai/pull/7337CompactOnRequest by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11422SystemPromptParts on each provider's operator channel with session.send() by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11379AnthropicModel.compact_messages() for on-demand native compaction by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11432BatchModelRequests capability and a get_wrapper_model capability hook by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10672GraphRun iterator and ToolManager context reference cycles by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11158Agent.run_stream name inference from keeping deps alive until garbage collection by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11219ModelResponse.provider_response_id by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11223deepseek-flash in DeepSeekProvider by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11249AgentStream event stream when a consumer leaves node.stream() early by @adtyavrdhn in https://github.com/pydantic/pydantic-ai/pull/11009thinking=False on OpenRouter's anthropic/claude-sonnet-5.5, which can't turn reasoning off by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11261on_change='append', and give an id-less URL MCPToolset its hostname as id by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10791dump_messages() by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11188too_many_requests stream errors for FallbackModel by @pydanty in https://github.com/pydantic/pydantic-ai/pull/11253google_proactive_audio work on the default Gemini provider with a separate v1alpha Live client by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11067load_messages() reloads a dumped history by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11192openai_previous_response_id chaining when a rewrite-mode instruction changes beside an on_change='append' block by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11331Memory snapshots in history once per changed run, so requests stay append-only by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10903thinking=False on OpenRouter routes that make reasoning mandatory by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11347ctx.emit() by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11157SpeechPart instead of splitting it at each delegated tool call by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11187CompactionPart with an empty summary instead of treating it as a compaction boundary by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11368AGUIAdapter.dump_messages injecting newlines between adjacent TextParts by @pydanty in https://github.com/pydantic/pydantic-ai/pull/8660AGUIAdapter.dump_messages does by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11391UIMessage of a split request its own id in Vercel AI dump_messages() by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11193FileUrl media_type win over the downloaded Content-Type, and treat binary/octet-stream as unknown by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11407ctx.emit() events into the realtime session event stream by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11369search_tools result as "No matching tools found" on Anthropic when its matches are no longer declared by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11411SummarizingCompaction and RelevanceCompaction transcript by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11370OpenAIChatModel and OpenAIResponsesModel by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11410reconnect policy by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11070ModelSettings: forced tool_choice raises on Gemini Live, parallel_tool_calls=False warns by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11066code_interpreter_call events by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11166ToolOutputLimits from reducing load_capability and search_tools returns, so loaded capabilities' tools unlock by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11446configuration_update system message by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11426GraphBuilder broadcast branches in .to() order instead of reversed, which can break replay of durable workflows paused across the upgrade by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11408capture_run_messages by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11413on_change='append' instruction blocks behind a second cache point after the static instructions by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11351ToolOutputLimits bands largest-first after combine, with the later band winning a shared threshold by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11453ToolReturn.content from CodeMode run_code to the model instead of silently dropping it by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11456ModelRetry raised after a BackgroundTools tool's step has ended toward max_retries by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11458SystemReminders on request-scoped system prompts and deprecate cache_ttl by @DouweM in https://github.com/pydantic/pydantic-ai/pull/9712SummarizingCompaction, SlidingWindowCompaction and RelevanceCompaction remove their load_capability exchange by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11394SSHWorkspaceBackend timeout and output-limit errors by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11479Shell persistent shell tool after a durable (DBOS) replay by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11457Full Changelog: https://github.com/pydantic/pydantic-ai/compare/v2.61.0...v2.62.0
configuration_update and Anthropic per-message output_config, recorded as ModelSettingsDeltaPart by @DouweM in #11111ModelRequestRejectedError, ModelAuthenticationError and ModelNotFoundError model error categories by @DouweM in #11081before_model_request by @DouweM in #11103tool_prefix by @DouweM in #10376SummarizingCompaction summary as a CompactionPart instead of a system prompt, and send provider-less CompactionParts to every model by @DouweM in #10573SystemPromptParts on each provider's operator channel with session.send() by @DouweM in #11379on_change='append', and give an id-less URL MCPToolset its hostname as id by @DouweM in #10791Memory snapshots in history once per changed run, so requests stay append-only by @DouweM in #10903ctx.emit() by @dsfaccini in #11157SpeechPart instead of splitting it at each delegated tool call by @DouweM in #11187ctx.emit() events into the realtime session event stream by @DouweM in #11369ModelSettings: forced tool_choice raises on Gemini Live, parallel_tool_calls=False warns by @DouweM in #11066configuration_update system message by @DouweM in #11426GraphBuilder broadcast branches in .to() order instead of reversed, which can break replay of durable workflows paused across the upgrade by @DouweM in #11408on_change='append' instruction blocks behind a second cache point after the static instructions by @DouweM in #11351SystemReminders on request-scoped system prompts and deprecate cache_ttl by @DouweM in #9712CodeMode(monty_sandbox=...), which takes a remote Monty URL or a function that builds the pool, and deprecate monty_sandbox_url by @adtyavrdhn in #11141configuration_update and Anthropic per-message output_config, recorded as ModelSettingsDeltaPart by @DouweM in #11111ModelRequestRejectedError, ModelAuthenticationError and ModelNotFoundError model error categories by @DouweM in #11081conversation_id_from_run_id from pydantic_ai.durable_exec for third-party durability engines by @DouweM in #11338child_workflow tool metadata in TemporalDurability by @YHallouard in #7329tool_choice subsets with allowed_tools on OpenAI Chat Completions to preserve the prompt cache by @DouweM in #10396before_model_request by @DouweM in #11103tool_prefix by @DouweM in #10376SummarizingCompaction summary as a CompactionPart instead of a system prompt, and send provider-less CompactionParts to every model by @DouweM in #10573tool_prefix to the MCP capability by @DouweM in #11364SubAgents run delegations as tasks the model can background, steer, answer and cancel, behind task_control by @mpfaffenberger in #10893Coder(task_control=) to turn on SubAgents task control by @mpfaffenberger in #11007StructuredContent user content, sent as its value to decision models by @DouweM in #10529RequestUsage.cost by @adtyavrdhn in #7337CompactOnRequest by @mpfaffenberger in #11422SystemPromptParts on each provider's operator channel with session.send() by @DouweM in #11379AnthropicModel.compact_messages() for on-demand native compaction by @DouweM in #11432BatchModelRequests capability and a get_wrapper_model capability hook by @DouweM in #10672GraphRun iterator and ToolManager context reference cycles by @dsfaccini in #11158Agent.run_stream name inference from keeping deps alive until garbage collection by @dsfaccini in #11219ModelResponse.provider_response_id by @dsfaccini in #11223deepseek-flash in DeepSeekProvider by @dsfaccini in #11249AgentStream event stream when a consumer leaves node.stream() early by @adtyavrdhn in #11009thinking=False on OpenRouter's anthropic/claude-sonnet-5.5, which can't turn reasoning off by @dsfaccini in #11261on_change='append', and give an id-less URL MCPToolset its hostname as id by @DouweM in #10791dump_messages() by @DouweM in #11188too_many_requests stream errors for FallbackModel by @pydanty in #11253google_proactive_audio work on the default Gemini provider with a separate v1alpha Live client by @DouweM in #11067load_messages() reloads a dumped history by @DouweM in #11192openai_previous_response_id chaining when a rewrite-mode instruction changes beside an on_change='append' block by @DouweM in #11331Memory snapshots in history once per changed run, so requests stay append-only by @DouweM in #10903thinking=False on OpenRouter routes that make reasoning mandatory by @DouweM in #11347ctx.emit() by @dsfaccini in #11157SpeechPart instead of splitting it at each delegated tool call by @DouweM in #11187CompactionPart with an empty summary instead of treating it as a compaction boundary by @DouweM in #11368AGUIAdapter.dump_messages injecting newlines between adjacent TextParts by @pydanty in #8660AGUIAdapter.dump_messages does by @DouweM in #11391UIMessage of a split request its own id in Vercel AI dump_messages() by @DouweM in #11193FileUrl media_type win over the downloaded Content-Type, and treat binary/octet-stream as unknown by @DouweM in #11407ctx.emit() events into the realtime session event stream by @DouweM in #11369search_tools result as "No matching tools found" on Anthropic when its matches are no longer declared by @DouweM in #11411SummarizingCompaction and RelevanceCompaction transcript by @DouweM in #11370OpenAIChatModel and OpenAIResponsesModel by @DouweM in #11410reconnect policy by @DouweM in #11070ModelSettings: forced tool_choice raises on Gemini Live, parallel_tool_calls=False warns by @DouweM in #11066code_interpreter_call events by @dsfaccini in #11166ToolOutputLimits from reducing load_capability and search_tools returns, so loaded capabilities' tools unlock by @mpfaffenberger in #11446configuration_update system message by @DouweM in #11426GraphBuilder broadcast branches in .to() order instead of reversed, which can break replay of durable workflows paused across the upgrade by @DouweM in #11408capture_run_messages by @DouweM in #11413on_change='append' instruction blocks behind a second cache point after the static instructions by @DouweM in #11351ToolOutputLimits bands largest-first after combine, with the later band winning a shared threshold by @mpfaffenberger in #11453ToolReturn.content from CodeMode run_code to the model instead of silently dropping it by @mpfaffenberger in #11456ModelRetry raised after a BackgroundTools tool's step has ended toward max_retries by @mpfaffenberger in #11458SystemReminders on request-scoped system prompts and deprecate cache_ttl by @DouweM in #9712SummarizingCompaction, SlidingWindowCompaction and RelevanceCompaction remove their load_capability exchange by @DouweM in #11394SSHWorkspaceBackend timeout and output-limit errors by @DouweM in #11479Shell persistent shell tool after a durable (DBOS) replay by @mpfaffenberger in #11457Full Changelog: v2.61.0...v2.62.0
Deprecate google_multi_speaker on GoogleRealtimeModel, which no Gemini Live model supports by @DouweM in https://github.com/pydantic/pydantic-ai/pull/…
<!-- Release notes generated using configuration in .github/release.yml at dd82a00ae3ef9a5f147405153eb877221d74d1c7 -->
thinking by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/10231thinking=False on models that can't turn thinking off by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11036NotImplementedError for a file in a CohereModel tool return instead of dropping it by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11114bool and int inputs on separate Literal decision branches in pydantic_graph by @dafyy321-pixel in https://github.com/pydantic/pydantic-ai/pull/9091google_multi_speaker on GoogleRealtimeModel, which no Gemini Live model supports by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11064auth functions in the hosted MCP capabilities by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10835CapabilityStore.load_active and stop overwriting unreadable manifests by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10731CapabilityStore lock per directory, and let runs cancel authoring by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10767BrowserUse prohibited lists browser-use would match by exact hostname, and retry when redacted schema keys collide by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10775LocalStack container inside async with agent:, stop service-list bypasses and aws configure writes to ~/.aws, and cap aws_cli output by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10828InputGuardrail redact the user's own prompt and close fail-open paths in guardrails by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10841CodeMode(dynamic_catalog=True) discoveries from rewriting the cached prompt by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10906auth functions in the hosted MCP capabilities by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10835TextPart by @adtyavrdhn in https://github.com/pydantic/pydantic-ai/pull/7434shell job logs with Shell.max_log_bytes by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10712kimi-k3.1-pro and kimi-k3.1-pro-highspeed model names by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11173transport= in create_async_httpx2_client() and give the httpx2 retry transports its pool limits by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11073RealtimeClientSecret.webrtc_url, which carries webrtcfilter=on on Azure OpenAI by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11076RealtimeModelProfile.supports_audio_output so text-only realtime models like Gemini's robotics streaming models default to text output by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11102ParameterTransports, DurableRunContext and run identity hooks by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11088reasoning that arrives without reasoning_details by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11104UserError instead of an AssertionError for an unconverted response SpeechPart in BedrockConverseModel by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11107NotImplementedError for a file in a CohereModel tool return instead of dropping it by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11114TOOL_CALL_ID_PREFIX instead of a hardcoded 'pyd_ai_' by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11109bool and int inputs on separate Literal decision branches in pydantic_graph by @dafyy321-pixel in https://github.com/pydantic/pydantic-ai/pull/9091NativeOutput on every FireworksProvider model by @pydanty in https://github.com/pydantic/pydantic-ai/pull/10157thinking by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/10231XaiModel native tool return parts by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11105OpenAIChatModel response with a non-standard finish_reason instead of raising UnexpectedModelBehavior by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11119InputGuardrail redact the user's own prompt and close fail-open paths in guardrails by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10841CapabilityStore.load_active and stop overwriting unreadable manifests by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10731Linear(id=None) uncombined like GitHub(id=None) by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10813ExaAgent poll leaves behind, and stop retrying Exa errors the model can't fix by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10840Shell commands over SSHWorkspaceBackend, and return a command too long to start to the model by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11004E2BSandbox probe timeouts and validate env and sandbox_timeout at construction by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10811ExaSearch and ExaAgent client on a new event loop, and build ExaAgent's once by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10848CapabilityStore lock per directory, and let runs cancel authoring by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10767ConversationSearch returning inherited history once per run, and clip excerpts around the matched term by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10749BrowserUse prohibited lists browser-use would match by exact hostname, and retry when redacted schema keys collide by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10775LocalStack container inside async with agent:, stop service-list bypasses and aws configure writes to ~/.aws, and cap aws_cli output by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10828InputGuardrail redaction also rewrite the history prompt when the model was sent a copy, and tighten the guardrails docs by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10852BrowserUse secrets in browse_web results by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10827arguments in HuggingFaceModel message history by @canblmz1 in https://github.com/pydantic/pydantic-ai/pull/11139RunCancelled across the durable tool-call boundary as a failed tool result by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11163XaiModel native tool call as a call part and one return part by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/11164gemini-2.5-flash-image, which rejects it by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11099logfire-cli wizard --task pydantic-ai by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11069OpenAIResponsesModel force a tool revealed mid-run on models before GPT-5.6 by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/10273CodeMode(dynamic_catalog=True) discoveries from rewriting the cached prompt by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10906SpeechPart by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11159defer_loading=True capability is loaded on Anthropic by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11091thinking=False on models that can't turn thinking off by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11036google_multi_speaker on GoogleRealtimeModel, which no Gemini Live model supports by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11064hang_up() for Azure OpenAI WebRTC calls by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11075session.started by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11115__pydantic_ai__ metadata out of sanitize_messages output, as the UI adapters do by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11089Full Changelog: https://github.com/pydantic/pydantic-ai/compare/v2.60.0...v2.61.0
thinking by @dsfaccini in #10231thinking=False on models that can't turn thinking off by @DouweM in #11036NotImplementedError for a file in a CohereModel tool return instead of dropping it by @dsfaccini in #11114bool and int inputs on separate Literal decision branches in pydantic_graph by @dafyy321-pixel in #9091google_multi_speaker on GoogleRealtimeModel, which no Gemini Live model supports by @DouweM in #11064auth functions in the hosted MCP capabilities by @mpfaffenberger in #10835CapabilityStore.load_active and stop overwriting unreadable manifests by @mpfaffenberger in #10731CapabilityStore lock per directory, and let runs cancel authoring by @mpfaffenberger in #10767BrowserUse prohibited lists browser-use would match by exact hostname, and retry when redacted schema keys collide by @mpfaffenberger in #10775LocalStack container inside async with agent:, stop service-list bypasses and aws configure writes to ~/.aws, and cap aws_cli output by @mpfaffenberger in #10828InputGuardrail redact the user's own prompt and close fail-open paths in guardrails by @mpfaffenberger in #10841CodeMode(dynamic_catalog=True) discoveries from rewriting the cached prompt by @DouweM in #10906auth functions in the hosted MCP capabilities by @mpfaffenberger in #10835TextPart by @adtyavrdhn in #7434shell job logs with Shell.max_log_bytes by @mpfaffenberger in #10712kimi-k3.1-pro and kimi-k3.1-pro-highspeed model names by @dsfaccini in #11173transport= in create_async_httpx2_client() and give the httpx2 retry transports its pool limits by @DouweM in #11073RealtimeClientSecret.webrtc_url, which carries webrtcfilter=on on Azure OpenAI by @DouweM in #11076RealtimeModelProfile.supports_audio_output so text-only realtime models like Gemini's robotics streaming models default to text output by @DouweM in #11102ParameterTransports, DurableRunContext and run identity hooks by @DouweM in #11088reasoning that arrives without reasoning_details by @dsfaccini in #11104UserError instead of an AssertionError for an unconverted response SpeechPart in BedrockConverseModel by @dsfaccini in #11107NotImplementedError for a file in a CohereModel tool return instead of dropping it by @dsfaccini in #11114TOOL_CALL_ID_PREFIX instead of a hardcoded 'pyd_ai_' by @dsfaccini in #11109bool and int inputs on separate Literal decision branches in pydantic_graph by @dafyy321-pixel in #9091NativeOutput on every FireworksProvider model by @pydanty in #10157thinking by @dsfaccini in #10231XaiModel native tool return parts by @dsfaccini in #11105OpenAIChatModel response with a non-standard finish_reason instead of raising UnexpectedModelBehavior by @dsfaccini in #11119InputGuardrail redact the user's own prompt and close fail-open paths in guardrails by @mpfaffenberger in #10841CapabilityStore.load_active and stop overwriting unreadable manifests by @mpfaffenberger in #10731Linear(id=None) uncombined like GitHub(id=None) by @mpfaffenberger in #10813ExaAgent poll leaves behind, and stop retrying Exa errors the model can't fix by @mpfaffenberger in #10840Shell commands over SSHWorkspaceBackend, and return a command too long to start to the model by @mpfaffenberger in #11004E2BSandbox probe timeouts and validate env and sandbox_timeout at construction by @mpfaffenberger in #10811ExaSearch and ExaAgent client on a new event loop, and build ExaAgent's once by @mpfaffenberger in #10848CapabilityStore lock per directory, and let runs cancel authoring by @mpfaffenberger in #10767ConversationSearch returning inherited history once per run, and clip excerpts around the matched term by @mpfaffenberger in #10749BrowserUse prohibited lists browser-use would match by exact hostname, and retry when redacted schema keys collide by @mpfaffenberger in #10775LocalStack container inside async with agent:, stop service-list bypasses and aws configure writes to ~/.aws, and cap aws_cli output by @mpfaffenberger in #10828InputGuardrail redaction also rewrite the history prompt when the model was sent a copy, and tighten the guardrails docs by @mpfaffenberger in #10852BrowserUse secrets in browse_web results by @mpfaffenberger in #10827arguments in HuggingFaceModel message history by @canblmz1 in #11139RunCancelled across the durable tool-call boundary as a failed tool result by @dsfaccini in #11163XaiModel native tool call as a call part and one return part by @dsfaccini in #11164gemini-2.5-flash-image, which rejects it by @DouweM in #11099logfire-cli wizard --task pydantic-ai by @DouweM in #11069OpenAIResponsesModel force a tool revealed mid-run on models before GPT-5.6 by @dsfaccini in #10273CodeMode(dynamic_catalog=True) discoveries from rewriting the cached prompt by @DouweM in #10906SpeechPart by @DouweM in #11159defer_loading=True capability is loaded on Anthropic by @DouweM in #11091thinking=False on models that can't turn thinking off by @DouweM in #11036google_multi_speaker on GoogleRealtimeModel, which no Gemini Live model supports by @DouweM in #11064hang_up() for Azure OpenAI WebRTC calls by @DouweM in #11075session.started by @DouweM in #11115__pydantic_ai__ metadata out of sanitize_messages output, as the UI adapters do by @DouweM in #11089Full Changelog: v2.60.0...v2.61.0
Deprecate RetryPromptPart in favor of ToolReturnPart(outcome='retried') and RetryFeedbackPart by @dsfaccini in #8094
ModelRateLimitError from embedding models by @DouweM in #10804usage_limits, deadline, run_id and conversation_id to ModelSelectionContext by @DouweM in #10856BedrockConverseModel's maxTokens to Claude's maximum output, streaming such requests behind the scenes by @DouweM in #10845RetryPromptPart in favor of ToolReturnPart(outcome='retried') and RetryFeedbackPart by @dsfaccini in #8094<harness_data> instead of user messages in SummarizingCompaction by @DouweM in #10833AskUser, ConversationSearch, Macroscope, ManagedPrompt, PlaywrightBrowser and PydanticAIDocs a stable default id, and append ManagedPrompt version changes by @DouweM in #10904MCP capability's default id from the server URL's hostname only, and reject two servers on one host without explicit ids by @DouweM in #10807MetadataMemoryStore and Memory(write_metadata=...) by @mpfaffenberger in #10892InstructionPart passed to a run's instructions opt into on_change='append', keyed run:<name> by @DouweM in #10789SummarizingCompaction(keep_user_messages=True) copies recognized after their request is merged by @DouweM in #10914instructions from untrusted history in sanitize_messages and server-managed UI adapters by @DouweM in #10897datetime and UUID, to the model unquoted so restored histories match the first run by @dsfaccini in #10488settings and profile in model equality and Model repr by @dsfaccini in #9474OpenAIDecisionsProvider and let OpenAIDecisionsModel take OpenAIProvider by @DouweM in #10951ToolAvailabilityDeltaPart by @DouweM in #9405send(text) runs the agent's tools by @DouweM in #10960StepPersistence snapshots, and fix SqliteMediaStore, DiskMediaStore, MongoMediaStore and S3MediaStore edge cases by @mpfaffenberger in #10834Advisor capabilities by @mpfaffenberger in #10726TextContent.metadata from user prompts in sanitize_messages by @DouweM in #10943ModelConnectionError and ModelTimeoutError for xAI gRPC connection failures and client timeouts by @DouweM in #11080system_prompt and seeded history's system prompts to realtime sessions by @DouweM in #11055SubAgents background task output as framed <task_output> data instead of in a SystemPromptPart by @DouweM in #10550BrowserUse secrets on allowlisted sites and out of tool results, and fix allowlist and file:// gaps by @mpfaffenberger in #10771FileSystem denied and read-only patterns case-insensitively, protect .git files, and bound reads and Python regex searches by @mpfaffenberger in #10825GitHub tools on readOnlyHint under read_only=True, reject 'oauth', and deprecate plain http:// URLs by @mpfaffenberger in #10815'max' to unified thinking effort levels by @dsfaccini in #10920DockerSandbox harness capability that runs the workspace in a local Docker or Podman container by @mpfaffenberger in #9341SmolSandboxBackend use a VM the application already has by @adtyavrdhn in #10882in_response_to_ids by @DouweM in #10792ModelSelectionContext by @DouweM in #10836ToolDefinition.unless_tool_kind and RunContext.tool_name_for_kind, and give instructions the current step's tools by @DouweM in #10673supports_audio_file_input and supported_document_media_types profile flags by @DouweM in #10476gh-aw copilot/ models on the github-copilot: model and declare detection-engine: copilot by @mpfaffenberger in #9318base_url and MOONSHOTAI_BASE_URL support to MoonshotAIProvider by @yanlinf in #10177FileSystem by @DouweM in #9410RelevanceCompaction, which keeps, reduces or drops each older history step by a judge model's ranking by @mpfaffenberger in #11026SmartSearch harness capability for plain-English code search with a pluggable judge model by @mpfaffenberger in #8977OpenAIDecisionsModel message history as native Responses input items by @DouweM in #11028FileUnderstanding capability by @DouweM in #8453MCPToolset take auth and headers as functions of the run context, so each run connects as its own user by @DouweM in #10846TemporalDurability via continue_as_new by @YHallouard in #7312lockdown to GitHub to ask GitHub's server to withhold public content from users without push access by @mpfaffenberger in #10822SetToolMetadata(background=...) having no effect on BackgroundTools by @adtyavrdhn in #10613ContentFilterError when Azure blocks a Responses API prompt without content_filter_result by @DouweM in #10805instructions recorded in seeded message_history by @dsfaccini in #10896BedrockConverseModel.request_stream() context exits, including on cancellation by @DouweM in #10900BedrockConverseModel message caching readable after a Claude message ending in a document by @DouweM in #10907output_type overloads for Agent.from_spec and Agent.from_file by @tjkuson in #10886CodeMode models to import asyncio before calling asyncio.gather by @adtyavrdhn in #10913openai_previous_response_id after a call to a tool with a tool_kind by @DouweM in #10931SystemPromptPart that no user message clears, as after a no_output retry on Anthropic by @DouweM in #10938InstructionBaselinePart by @DouweM in #10911ClearToolResults and DeduplicateFileReads by @DouweM in #10932VercelAIAdapter round-trip by @DouweM in #10933pydantic_ai.tool.retry.* attributes on argument-validation failure spans by @DouweM in #10934CodeMode tool calls, and a first model request that raises, until a parallel InputGuardrail decides by @dsfaccini in #10365AWSLambdaDurability by @mpfaffenberger in #10730Grain's write tools and treat Grain(id=None) as the default grain id by @mpfaffenberger in #10801AskUserAnsweredEvent with failed=True when the AskUser answerer raises by @mpfaffenberger in #10725played_audio_bytes and wait_for_playback() instead of a UserError once it ended the audio stream by @DouweM in #10940_framing.wrap, so a </system> in an instruction or tool name can't end its block by @DouweM in #10547ModelRetry and ToolFailed from execute hooks on BackgroundTools calls to the model by @mpfaffenberger in #10830AskUser questions and document its timeout, concurrency and tracing behavior by @mpfaffenberger in #10755RunError instead of denying silently by @dsfaccini in #7038ToolSearch(strategy='bm25') on the legacy Bedrock AsyncAnthropicBedrock client by @dsfaccini in #10340reasoning_content on DeepSeekProvider tool calls whenever the request thinks by @dsfaccini in #10350TemplateVariable and fall back to the code default on invalid templates in ManagedPrompt, and correct its docs by @mpfaffenberger in #10821ModalSandboxBackend and log a sandbox created after every caller was cancelled by @mpfaffenberger in #10816session_output_overflow as non-recoverable without a reconnect policy by @DouweM in #10947usage.iterations pass by @dsfaccini in #10363RealtimeTurnCompleteEvent once for a realtime turn that ends because a tool raised by @DouweM in #10965retry-after-ms in wait_retry_after by @DouweM in #11065answer_webrtc_offer from rejecting the profile's audio sample rates by @DouweM in #11049tool_search call so the model gets its result instead of searching again by @DouweM in #11058input_transcription_model='auto' on OpenAI GPT-Live by @DouweM in #11047Macroscope work under durable execution and return the findings of a review the CLI stopped early by @mpfaffenberger in #10817LogfireMCP(id=None) as the default id, accept client with auth='', and flag untrusted telemetry in its docs by @mpfaffenberger in #10818anthropic 2.x: build message params as dicts and keep Bedrock's anthropic_version by @DouweM in #11077prefect>=3.9.0 so the prefect extra imports on current FastAPI by @DouweM in #11071ModelRequest fields when WarnNearLimits strips legacy warnings by @mpfaffenberger in #10776ApprovalReviewer and ToolCallJudge the arguments as written, and correct the approval reviewer docs by @mpfaffenberger in #10727BubblewrapWorkspace by @mpfaffenberger in #10746FallbackModel response handlers when continuing a suspended response by @DouweM in #11085AgentPlugin by @DouweM in #11082thinking to reasoning.effort for OpenAI GPT-5.6 and GPT-6 on Bedrock Converse by @rscholz98 in #10993SystemOneModel backends like vLLM by @mpfaffenberger in #11092SlidingWindowCompaction and SummarizingCompaction histories sendable after tool retries and keep_messages=0, and warn on an invalid summary_prompt by @mpfaffenberger in #10759E2BSandbox commands on templates without ps by @mpfaffenberger in #10809auth='oauth' and describe only the selected products in GoogleWorkspace by @mpfaffenberger in #10800prefixItems and string length bounds in OpenAI strict mode by @adtyavrdhn in #7333E2BSandbox raise a clear UserError on Trio before any E2B request by @mpfaffenberger in #10810CodeMode speculation adopting reads that predate a write, scan eager streams that split \n, and reject out-of-range resource_limits by @mpfaffenberger in #10764Note truncated.
<!-- Release notes generated using configuration in .github/release.yml at 7bccbb8096be8249ebf0ed1d8f01a6fae7b20acb -->
ModelRateLimitError from embedding models by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10804usage_limits, deadline, run_id and conversation_id to ModelSelectionContext by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10856BedrockConverseModel's maxTokens to Claude's maximum output, streaming such requests behind the scenes by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10845RetryPromptPart in favor of ToolReturnPart(outcome='retried') and RetryFeedbackPart by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/8094<harness_data> instead of user messages in SummarizingCompaction by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10833AskUser, ConversationSearch, Macroscope, ManagedPrompt, PlaywrightBrowser and PydanticAIDocs a stable default id, and append ManagedPrompt version changes by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10904MCP capability's default id from the server URL's hostname only, and reject two servers on one host without explicit ids by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10807MetadataMemoryStore and Memory(write_metadata=...) by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10892InstructionPart passed to a run's instructions opt into on_change='append', keyed run:<name> by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10789SummarizingCompaction(keep_user_messages=True) copies recognized after their request is merged by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10914instructions from untrusted history in sanitize_messages and server-managed UI adapters by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10897datetime and UUID, to the model unquoted so restored histories match the first run by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/10488settings and profile in model equality and Model repr by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/9474OpenAIDecisionsProvider and let OpenAIDecisionsModel take OpenAIProvider by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10951ToolAvailabilityDeltaPart by @DouweM in https://github.com/pydantic/pydantic-ai/pull/9405send(text) runs the agent's tools by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10960StepPersistence snapshots, and fix SqliteMediaStore, DiskMediaStore, MongoMediaStore and S3MediaStore edge cases by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10834Advisor capabilities by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10726TextContent.metadata from user prompts in sanitize_messages by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10943ModelConnectionError and ModelTimeoutError for xAI gRPC connection failures and client timeouts by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11080system_prompt and seeded history's system prompts to realtime sessions by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11055SubAgents background task output as framed <task_output> data instead of in a SystemPromptPart by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10550BrowserUse secrets on allowlisted sites and out of tool results, and fix allowlist and file:// gaps by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10771FileSystem denied and read-only patterns case-insensitively, protect .git files, and bound reads and Python regex searches by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10825GitHub tools on readOnlyHint under read_only=True, reject 'oauth', and deprecate plain http:// URLs by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10815'max' to unified thinking effort levels by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/10920DockerSandbox harness capability that runs the workspace in a local Docker or Podman container by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/9341SmolSandboxBackend use a VM the application already has by @adtyavrdhn in https://github.com/pydantic/pydantic-ai/pull/10882in_response_to_ids by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10792ModelSelectionContext by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10836ToolDefinition.unless_tool_kind and RunContext.tool_name_for_kind, and give instructions the current step's tools by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10673supports_audio_file_input and supported_document_media_types profile flags by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10476gh-aw copilot/ models on the github-copilot: model and declare detection-engine: copilot by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/9318base_url and MOONSHOTAI_BASE_URL support to MoonshotAIProvider by @yanlinf in https://github.com/pydantic/pydantic-ai/pull/10177FileSystem by @DouweM in https://github.com/pydantic/pydantic-ai/pull/9410RelevanceCompaction, which keeps, reduces or drops each older history step by a judge model's ranking by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11026SmartSearch harness capability for plain-English code search with a pluggable judge model by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/8977OpenAIDecisionsModel message history as native Responses input items by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11028FileUnderstanding capability by @DouweM in https://github.com/pydantic/pydantic-ai/pull/8453MCPToolset take auth and headers as functions of the run context, so each run connects as its own user by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10846TemporalDurability via continue_as_new by @YHallouard in https://github.com/pydantic/pydantic-ai/pull/7312lockdown to GitHub to ask GitHub's server to withhold public content from users without push access by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10822SetToolMetadata(background=...) having no effect on BackgroundTools by @adtyavrdhn in https://github.com/pydantic/pydantic-ai/pull/10613ContentFilterError when Azure blocks a Responses API prompt without content_filter_result by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10805instructions recorded in seeded message_history by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/10896BedrockConverseModel.request_stream() context exits, including on cancellation by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10900BedrockConverseModel message caching readable after a Claude message ending in a document by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10907output_type overloads for Agent.from_spec and Agent.from_file by @tjkuson in https://github.com/pydantic/pydantic-ai/pull/10886CodeMode models to import asyncio before calling asyncio.gather by @adtyavrdhn in https://github.com/pydantic/pydantic-ai/pull/10913openai_previous_response_id after a call to a tool with a tool_kind by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10931SystemPromptPart that no user message clears, as after a no_output retry on Anthropic by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10938InstructionBaselinePart by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10911ClearToolResults and DeduplicateFileReads by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10932VercelAIAdapter round-trip by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10933pydantic_ai.tool.retry.* attributes on argument-validation failure spans by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10934CodeMode tool calls, and a first model request that raises, until a parallel InputGuardrail decides by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/10365AWSLambdaDurability by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10730Grain's write tools and treat Grain(id=None) as the default grain id by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10801AskUserAnsweredEvent with failed=True when the AskUser answerer raises by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10725played_audio_bytes and wait_for_playback() instead of a UserError once it ended the audio stream by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10940_framing.wrap, so a </system> in an instruction or tool name can't end its block by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10547ModelRetry and ToolFailed from execute hooks on BackgroundTools calls to the model by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10830AskUser questions and document its timeout, concurrency and tracing behavior by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10755RunError instead of denying silently by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/7038ToolSearch(strategy='bm25') on the legacy Bedrock AsyncAnthropicBedrock client by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/10340reasoning_content on DeepSeekProvider tool calls whenever the request thinks by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/10350TemplateVariable and fall back to the code default on invalid templates in ManagedPrompt, and correct its docs by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10821ModalSandboxBackend and log a sandbox created after every caller was cancelled by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10816session_output_overflow as non-recoverable without a reconnect policy by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10947usage.iterations pass by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/10363RealtimeTurnCompleteEvent once for a realtime turn that ends because a tool raised by @DouweM in https://github.com/pydantic/pydantic-ai/pull/10965retry-after-ms in wait_retry_after by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11065answer_webrtc_offer from rejecting the profile's audio sample rates by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11049tool_search call so the model gets its result instead of searching again by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11058input_transcription_model='auto' on OpenAI GPT-Live by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11047Macroscope work under durable execution and return the findings of a review the CLI stopped early by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10817LogfireMCP(id=None) as the default id, accept client with auth='', and flag untrusted telemetry in its docs by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10818anthropic 2.x: build message params as dicts and keep Bedrock's anthropic_version by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11077prefect>=3.9.0 so the prefect extra imports on current FastAPI by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11071ModelRequest fields when WarnNearLimits strips legacy warnings by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10776ApprovalReviewer and ToolCallJudge the arguments as written, and correct the approval reviewer docs by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10727BubblewrapWorkspace by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10746FallbackModel response handlers when continuing a suspended response by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11085AgentPlugin by @DouweM in https://github.com/pydantic/pydantic-ai/pull/11082thinking to reasoning.effort for OpenAI GPT-5.6 and GPT-6 on Bedrock Converse by @rscholz98 in https://github.com/pydantic/pydantic-ai/pull/10993SystemOneModel backends like vLLM by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/11092SlidingWindowCompaction and SummarizingCompaction histories sendable after tool retries and keep_messages=0, and warn on an invalid summary_prompt by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10759E2BSandbox commands on templates without ps by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10809auth='oauth' and describe only the selected products in GoogleWorkspace by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10800prefixItems and string length bounds in OpenAI strict mode by @adtyavrdhn in https://github.com/pydantic/pydantic-ai/pull/7333E2BSandbox raise a clear UserError on Trio before any E2B request by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10810CodeMode speculation adopting reads that predate a write, scan eager streams that split \n, and reject out-of-range resource_limits by @mpfaffenberger in https://github.com/pydantic/pydantic-ai/pull/10764Full Changelog: https://github.com/pydantic/pydantic-ai/compare/v2.59.0...v2.60.0
Nothing was removed, but code that reads message history behaves differently:
RetryPromptPart][pydantic_ai.messages.RetryPromptPart]. A tool retry is that call's [ToolReturnPart][pydantic_ai.messages.ToolReturnPart] with outcome='retried', and a retry that answers no tool call is a [RetryFeedbackPart][pydantic_ai.messages.RetryFeedbackPart]. A stored 'retry-prompt' part loads as one of those too, so isinstance(part, RetryPromptPart) matches nothing, and code that counts tool errors that way reports none without an error or warning. Replace that check with [is_retry(part)][pydantic_ai.messages.is_retry], which matches both new parts, or on a release that predates it, with isinstance(part, RetryFeedbackPart) or (isinstance(part, ToolReturnPart) and part.outcome == 'retried'); Finding tool errors in history shows it in use, and how to also find tool calls that failed. The deprecation entry lists the other changes.One column per month.