Skip to content

Releases

v2.69.0 (2026-11-16)

Choose a tag to compare

@dsfaccini dsfaccini released this 17 Nov 01:03
25b102d

What's Changed

⚠️ Compatibility Notes

  • Fold RequestInput into HandleDeferredToolCalls, whose handler can take a resume token by @DouweM in #13285
  • Support Gemini 4 Hydrogen (gemini-4-hydrogen) on GoogleModel by @dsfaccini in #13299
  • Generalize response-scoped tool call IDs into the tool_call_ids_are_response_scoped profile flag by @DouweM in #13128
  • Add service_tier support to GroqModel by @dsfaccini in #13521
  • Detect loops in streamed thinking, text and tool call arguments in DetectLoops by @mpfaffenberger in #12222
  • Make Advisor accept native= and local= like the other NativeOrLocalTool capabilities, deprecating mode by @adtyavrdhn in #13149
  • Treat nested CodeMode ToolFailed calls as failed results by @pydanty in #13076
  • Keep settled sub-agent task histories in their saved files instead of in every DelegationTasks owner's memory by @DouweM in #13221
  • Raise ModelQuotaExceededError for Anthropic's tier spend cap and user-set spend limits by @dsfaccini in #13269
  • Send structured tool return items next to a file to the model as JSON, not as escaped JSON strings by @dsfaccini in #13271
  • Raise a gave-up transport status or connection error from Google embeddings, image generation and GoogleModel.count_tokens as ModelAPIError by @dsfaccini in #13312
  • Raise a Gemini API error from GoogleImageGenerationModel.generate as its error category, like ModelRateLimitError by @dsfaccini in #13335
  • Raise an error status a retrying transport gave up on during a streamed Gemini long decode as ModelHTTPError by @dsfaccini in #13356
  • Raise a gave-up transport status from OpenAI and Cohere embeddings, OpenAI image generation and GoogleModel batch methods as ModelHTTPError by @dsfaccini in #13327
  • Cache opaque Prefect run IDs across retries by @dsfaccini in #11896
  • Reject a TextOutput function that takes a model, dataclass or TypedDict when the agent is created by @dsfaccini in #13422
  • Keep the body of an error status that HTTPX2TenacityTransport and AsyncHTTPX2TenacityTransport gave up on by @dsfaccini in #13411
  • Raise OpenAI API and connection errors from OpenAIImageGenerationModel.generate as their error category, like ModelRateLimitError by @dsfaccini in #13433
  • Raise realtime connection errors with an HTTP status as their error category, like ModelRateLimitError by @dsfaccini in #13436
  • Keep Bedrock GPT-5.6 and GPT-6 encrypted reasoning out of Claude-on-Bedrock requests by @pydanty in #12943
  • Raise ModelContextWindowExceededError for Mistral 3059 prompt-too-long errors, not for other 3051 invalid-argument errors by @pydanty in #12882
  • Raise a status a retrying transport gave up on as the error category its body names, like ModelQuotaExceededError by @dsfaccini in #13485
  • Stop counting OpenRouter's is_byok flag as a usage detail by @pydanty in #13025
  • Map Gemini IMAGE_RECITATION and TOO_MANY_TOOL_CALLS finish reasons to content_filter and error by @pydanty in #10918
  • Make TestModel integers satisfy mixed inclusive and exclusive bounds, and fractional bounds by @pydanty in #13298
  • Add minimax_model_profile and use it for MiniMax models on Bedrock, OpenAICompatibleProvider, LiteLLMProvider and TogetherProvider by @pydanty in #12693
  • Map Mistral Large 4 thinking levels to its low/medium/high reasoning_effort values by @pydanty in #12876
  • Apply FunctionToolset's strict, sequential, defer_loading and include_return_schema defaults to Tool instances by @pydanty in #12572
  • Make TestModel streamed usage match run() for tool calls and text by @pydanty in #11877
  • Run the token count count_tokens_before_request asks for in its own durable unit by @dsfaccini in #13460
  • Skip a failing TieredCompaction tier on model errors, and let tiers omit their trigger by @mpfaffenberger in #13660
  • Rename harness subagents, filesystem and spend packages to their capability names and align store, ACP* and *Error class names by @mpfaffenberger in #13401
  • Rename the DockerSandbox harness capability to DockerContainer, keeping DockerSandbox as a deprecated alias by @adtyavrdhn in #13245
  • Rename BubblewrapSandbox to BubblewrappedWorkspace and clarify its docs by @mpfaffenberger in #13179

🚀 Features

  • Trim Coder's default instructions and add Coder(autonomy=...) by @mpfaffenberger in #12542
  • Fold RequestInput into HandleDeferredToolCalls, whose handler can take a resume token by @DouweM in #13285
  • Support Gemini 4 Hydrogen (gemini-4-hydrogen) on GoogleModel by @dsfaccini in #13299
  • Add is_retry to find retries in message history and list the v2.60.0 RetryPromptPart change as breaking by @DouweM in #11404
  • Move benchmark-only Coder instructions into opt-in BenchmarkInstructions by @mpfaffenberger in #13395
  • Generalize response-scoped tool call IDs into the tool_call_ids_are_response_scoped profile flag by @DouweM in #13128
  • Add peer messaging with Peers in pydantic-ai-harness by @dsfaccini in #11671
  • Add an anthropic_inference_geo setting for Anthropic data residency by @rscholz98 in #12769
  • Add opt-in recursive node registration with GraphBuilder.add_reachable() by @dsfaccini in #12706
  • Add WindowsWorkspace, a PowerShell workspace for Windows hosts by @mpfaffenberger in #13475
  • Add service_tier support to GroqModel by @dsfaccini in #13521
  • Add KubernetesSandbox for kubernetes-sigs/agent-sandbox pods by @adtyavrdhn in #10612
  • Add E2BSandbox(network=), passing E2B's network options through to a new sandbox by @adtyavrdhn in #13587
  • Add BoxdSandbox(secrets=), naming the boxd secrets a new machine gets by @adtyavrdhn in #13589
  • Add pin(source=) so harness compaction re-injects tool- and model-written pins as framed data, not the user's words by @mpfaffenberger in #13615
  • Let Advisor.max_uses take a callable so a consultation budget can span several runs by @mpfaffenberger in #13617
  • Add DynamicWorkflowToolset.create_executor so subclasses can customize how sub-agent calls are scheduled by @mpfaffenberger in #13623
  • Detect loops in streamed thinking, text and tool call arguments in DetectLoops by @mpfaffenberger in #12222
  • Add openai_safety_identifier model setting by @pydanty in #12258
  • Add DaytonaSandbox(secrets=), mounting Daytona secrets in a new sandbox by @adtyavrdhn in #13590
  • Add OpenSandbox(network_policy=) and OpenSandboxBackend(credential_proxy=), passing OpenSandbox's egress options through by @adtyavrdhn in #13591
  • Wait a model error's retry_after durably before DBOS retries a model request, via a shared model_error_retry_after helper by @DouweM in #13270
  • Add CapabilityCreation.load_active(workspace), which refuses a store directory below 'operator' authority before loading it by @mpfaffenberger in #13641
  • Report pydantic-ai-harness in MongoDB handshake driver metadata by @pydanty in http...
Read more

CLAI2 bleeding

CLAI2 bleeding Pre-release
Pre-release

Choose a tag to compare

@github-actions github-actions released this 06 Oct 01:25
65471dd

Sdists rebuilt from main for /update on the CLAI bleeding channel. Not a release.

v2.68.0 (2026-11-11)

Choose a tag to compare

@dsfaccini dsfaccini released this 12 Nov 01:56
d8bcca9

What's Changed

⚠️ Compatibility Notes

  • Note text that looks like an application frame in user and tool content instead of escaping it by @DouweM in #11813
  • Add OutputTokenLimitExceeded, and fall back to the next model in ModelFallback when a response is cut off at max_tokens by @DouweM in #12407
  • Ask for approvals and answers inside the run with RequestInput, put AskUser on the deferred pipeline, and mark deferred calls with kind and answered_by by @DouweM in #12757
  • Add DelegationTasks.discover() and stop one damaged saved task from blocking DelegationTasks.opened() by @mpfaffenberger in #10140
  • Require workspace backends to leave detached processes running, and check it in WorkspaceBackendSuite by @mpfaffenberger in #13070
  • Run PromptInjectionDefender semantic classification as durable operations so it works on Temporal by @mpfaffenberger in #12424
  • Give sub-agent task-control events the Event suffix and past-participle names, with agent_name and parent_task_id fields, keeping the old names as deprecated aliases by @mpfaffenberger in #13074
  • Spell out abbreviated identifiers in pydantic_ai_harness, renaming num_results to max_results and max_duration_secs to max_duration_seconds by @mpfaffenberger in #13075
  • Rename harness CapabilityEvent fields to *_count, *_name, duration_ms and previous_item, keeping the old names as deprecated aliases by @mpfaffenberger in #13068
  • Rename harness CapabilityEvent outliers to <Subject><PastParticiple>Event and add STEP_PERSISTENCE_EVENTS by @mpfaffenberger in #13065
  • Add the RefineOutput harness capability by @mpfaffenberger in #13120

🚀 Features

  • Load and save UI adapter conversations through a ConversationStore with conversation_store by @DouweM in #12326
  • Use the AI Gateway's streamed cost for GoogleModel and BedrockConverseModel streams by @ddanielcruzz in #13168
  • Support running SubAgents delegations to durable delegates as Temporal and DBOS child workflows by @mpfaffenberger in #12516
  • Add the RefineOutput harness capability by @mpfaffenberger in #13120
  • Add a per-command cwd to Workspace.run by @mpfaffenberger in #12642
  • Require workspace backends to leave detached processes running, and check it in WorkspaceBackendSuite by @mpfaffenberger in #13070
  • Add DelegationTasks.discover() and stop one damaged saved task from blocking DelegationTasks.opened() by @mpfaffenberger in #10140
  • Bound the wait for the detached process to start in the WorkspaceBackendSuite reattach rule by @mpfaffenberger in #13081
  • Ask for approvals and answers inside the run with RequestInput, put AskUser on the deferred pipeline, and mark deferred calls with kind and answered_by by @DouweM in #12757
  • Add OutputTokenLimitExceeded, and fall back to the next model in ModelFallback when a response is cut off at max_tokens by @DouweM in #12407
  • Allow a CancellationToken fired from a workflow signal to cancel a Temporal durable run by @DouweM in #7210
  • Note text that looks like an application frame in user and tool content instead of escaping it by @DouweM in #11813
  • Add a GoogleDecisionsModel backend for the Gemini Decisions API by @dsfaccini in #13181

🐛 Bug Fixes

  • Resolve vendor/model ids through one case-insensitive family map on OpenAICompatibleProvider, LiteLLMProvider and TogetherProvider by @DouweM in #12743
  • Map unified thinking to reasoning_effort for Z.AI GLM and DeepSeek V3.2 on Bedrock by @EdAbati in #11491
  • Honor Permissions remembered approvals inside durable execution by @mpfaffenberger in #12410
  • Make AnthropicModel.count_tokens send the request's betas and raise its all-deferred tools UserError by @dsfaccini in #13098
  • Drop the account-age claim from AnthropicStaleThinkingBlockWarning, since Claude Fable 5.2 and Mythos 5.2 bind thinking blocks on every account by @dsfaccini in #13155
  • Run PromptInjectionDefender semantic classification as durable operations so it works on Temporal by @mpfaffenberger in #12424
  • Carry ModelResponse.canonical_system through the Vercel AI and AG-UI event streams by @DouweM in #13134
  • Lower a max_tokens above a Claude model's maximum output to it instead of sending it by @DouweM in #12427
  • Report when TieredCompaction can't reach target_tokens because its tiers keep more than the target by @DouweM in #12650
  • Summarize the tail messages SummarizingCompaction(keep_user_messages=True) copies displace instead of dropping them by @DouweM in #12652
  • Complete a Gemini Live turn that ends silently after a tool result by @DouweM in #13126

📦 Dependencies

  • Raise the genai-prices floor to 0.2.3 so Anthropic fast mode responses are priced at fast mode rates by @dsfaccini in #13104

New Contributors

Full Changelog: v2.67.0...v2.68.0

v2.67.0 (2026-11-10)

Choose a tag to compare

@dsfaccini dsfaccini released this 10 Nov 23:11
dd8489a

What's Changed

⚠️ Compatibility Notes

  • Support SubAgents(dynamic_agents=...) under durable execution by @mpfaffenberger in #12418
  • Strip every default-ignorable character before ArtifactScanner phrase matching and flag hidden variation selectors by @mpfaffenberger in #11475
  • Load one instruction file per directory and nested files by default in RepoContext, add walk_up, and have Coder build on them by @DouweM in #11691
  • Cap inlined nested instructions at 32,000 characters per conversation by default with RepoContext(nested_max_chars=...) by @DouweM in #11699
  • Send documents of any media type mimetypes knows, with the format taken from it by @arjun-krishna1 in #12365
  • Move provider-specific native-tool fields to provider_settings by @adtyavrdhn in #7807
  • Add framing_announcement setting and only announce the <system> voice tag once a conversation needs it by @DouweM in #12661
  • Surface nested RepoContext instruction files when the model writes or edits in a directory by @DouweM in #11701
  • Record InputGuardrail, OutputGuardrail and ToolGuardrail verdicts as durable operations by @mpfaffenberger in #12428
  • Refuse a CapabilityCreation run when directory is below 'operator' authority by @DouweM in #12558
  • Stop continuing an OpenAI response without a history_digest through openai_previous_response_id by @DouweM in #12575
  • Replay Prefect tasks on flow retry when the message history holds binary content by @sangkyoonnam in #12188
  • Reject CodeMode tools named after Python builtins, which the sandbox called instead of the tool by @adtyavrdhn in #12698
  • Break chat-template special tokens in content on open-model routes, and restore them in model output, with ModelProfile.may_tokenize_special_tokens_in_content by @DouweM in #12556
  • Use the engine's codec and serialization_failure for durable capability results by @maksimzayats in #11057
  • Send FileSearchTool.max_num_results to OpenAI Responses by @adtyavrdhn in #12728
  • Raise UserError 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 #12958
  • Record a StepPersistence tool call that ends in ModelRetry or ToolFailed as failed instead of leaving it started by @dsfaccini in #12638
  • Raise UserError from StreamedRunResult.result when a run_stream() run stops while finishing by @dsfaccini in #13087
  • Run CodeMode remote sandboxes over gRPC with pydantic-monty 1.5.0 by @dmontagu in #12673
  • Replace httpx with httpx2 in pydantic-ai-harness by @Kludex in #13026
  • Curate KnownModelName and the provider model-name literals to the models a user would pick by @dsfaccini in #12817

🚀 Features

  • Add dequeue and set_enqueue_priority to change a message queued with enqueue before delivery by @adtyavrdhn in #12345
  • Support SubAgents(dynamic_agents=...) under durable execution by @mpfaffenberger in #12418
  • Run tool calls as DBOS child workflows via child_workflow tool metadata in DBOSDurability by @mpfaffenberger in #12445
  • Checkpoint TrajectoryJudge evaluations as a durable operation when the judge has an id by @mpfaffenberger in #12414
  • Identify Azure Durable Task workers with pydantic-ai in their gRPC user agent by @tamirdresher in #12493
  • Add client to ModalSandbox and ModalSandboxBackend by @adtyavrdhn in #12501
  • Strip every default-ignorable character before ArtifactScanner phrase matching and flag hidden variation selectors by @mpfaffenberger in #11475
  • Add client to E2BSandbox and E2BSandboxBackend by @adtyavrdhn in #12500
  • Add per-path workspace authority with Workspace.authority_of by @DouweM in #12473
  • Add openai_include_unmapped_response_data to keep Chat Completions response data Pydantic AI doesn't map by @DouweM in #12034
  • Load one instruction file per directory and nested files by default in RepoContext, add walk_up, and have Coder build on them by @DouweM in #11691
  • Cap inlined nested instructions at 32,000 characters per conversation by default with RepoContext(nested_max_chars=...) by @DouweM in #11699
  • Support OpenAI fast and ultrafast service tiers in openai_service_tier by @adtyavrdhn in #12527
  • Send documents of any media type mimetypes knows, with the format taken from it by @arjun-krishna1 in #12365
  • Run DBOS child-workflow tools in parallel when a workspace is attached by @mpfaffenberger in #12522
  • Add the KeenableSearch capability for keyless web research by @ilya-bogin-keenable in #9079
  • Add gateway/bedrock:global.moonshotai.kimi-k3 and gateway/bedrock:global.moonshotai.kimi-k3.1 to KnownModelName by @dsfaccini in #12869
  • Add provider-specific settings to native tools by @adtyavrdhn in #7633
  • Support Anthropic web_search_20260318 and web_fetch_20260318 via provider_settings by @adtyavrdhn in #12610
  • Move provider-specific native-tool fields to provider_settings by @adtyavrdhn in #7807
  • Support SubAgents(task_control=...) under Temporal and DBOS durability by @mpfaffenberger in #12521
  • Support OpenAI's gpt-image-2.6-flare and gpt-image-2.6-sunburst image models by @dsfaccini in #12820
  • Add Claude Fable 5.2 (claude-fable-5-2) and Claude Mythos 5.2 (claude-mythos-5-2) support by @dsfaccini in #12970
  • List Claude Fable 5.2 on Bedrock (global.anthropic.claude-fable-5-2) and send per-message effort to it on OpenRouter by @dsfaccini in #13037
  • Support Gemini 4 Carbon (gemini-4-carbon) on GoogleModel by @dsfaccini in #13057
  • Add framing_announcement setting and only announce the <system> voice tag once a conversation needs it by @DouweM in #12661
  • Surface nested RepoContext instruction files when the model writes or edits in a directory by @DouweM in #11701

🐛 Bug Fixes

Read more

v2.66.0 (2026-11-04)

Choose a tag to compare

@adtyavrdhn adtyavrdhn released this 04 Nov 11:51
7d1d3f0

What's Changed

🛡️ Security

  • 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 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).
  • GHSA-mfx5-pc32-32m8 (high): The 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.
  • GHSA-5fj9-5pjw-vpwc (medium): The cloud-metadata blocklist that URL downloads check did not include Tencent Cloud's instance metadata address 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).
  • GHSA-jp94-mmr3-2pvv (medium): With OpenTelemetry instrumentation enabled for embedding models, embedding spans recorded the 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.
  • GHSA-7494-925c-rrxv (medium): For models whose profile requires tool schemas without $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).
  • GHSA-484x-x9m4-62cj (medium): Decision models (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).
  • GHSA-r879-rxgj-3g75 (medium): The Vercel AI and AG-UI adapters took time that grows with the square of the number of parts to load a submitted message, on the event loop, so a single request could delay other work in the same process. You are affected if you pass client-submitted conversations to VercelAIAdapter or AGUIAdapter. Fixed in v2.66.0 and v1.107.10 (pydantic-ai, pydantic-ai-slim).
  • GHSA-4cj4-hgwf-4hc4 (medium): When message history was serialized and loaded again, a tool result containing a JSON object in the serialized shape of one of Pydantic AI's file types was rebuilt as that file. A third party who controls data a tool returns could use this to make the application download a URL on its local network, or make the model provider read an 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).
  • GHSA-5ppx-9jq6-6pmw (low): 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.
  • GHSA-mgmv-r532-pg36 (medium): An 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.

⚠️ Compatibility Notes

  • Call Anthropic's 20260209 web tools directly on the Anthropic API when citations are requested by @dsfaccini in #12278
  • Apply FunctionToolset(requires_approval=True) to Tool instances passed to the toolset by @dsfaccini in #12378
  • Re-resolve realtime session instructions before each model turn and deliver their changes as appends by @DouweM in #11930
  • Recognize SummarizingCompaction(keep_user_messages=True) copies only by the marker on their text by @DouweM in #12316

🚀 Features

  • Add OpenSandbox for self-hosted agent workspaces by @adtyavrdhn in #11841
  • Add Azure Durable Task durability to pydantic_ai_harness by @DouweM in #11332
  • Add Pixeltable tools and memory storage to Harness by @pierrebrunelle in #9369
  • Add RenderSandbox to run agent workspaces in Render sandboxes by @ojusave in #10238
  • Add include_instructions to the MCP capability so server instructions can be left out of the agent's instructions by @DouweM in #12269
  • Re-resolve realtime session instructions before each model turn and deliver their changes as appends by @DouweM in #11930
  • Report xAI Grok Voice speech timing as segment timestamps with output_speech_timestamps by @DouweM in #11782
  • Add a retries field to SummarizingCompaction by @gouyu-hou in #12315
  • Add Workspace.authority and AuthorityWorkspace for workspaces whose files the operator does not control by @adtyavrdhn in #11836

🐛 Bug Fixes

  • Stop ModalSandbox failing with AttributeError on Modal 1.6, which removed FilesystemExecutionError by @adtyavrdhn in #12247
  • Call Anthropic's 20260209 web tools directly on the Anthropic API when citations are requested by @dsfaccini in #12278
  • Add streaming support to CohereModel by @Julien-ser in #6100
  • Release run dependencies after exceptions by @dsfaccini in #11255
  • Preserve extra keys in tool_search and load_capability message history by @pydanty in #7933
  • Inline $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 #12138
  • Keep RepoContext from following symlinks out of the working directory or to files that aren't instruction files by @DouweM in #12044
  • Reject approvals for tool calls a server-held history isn't waiting on under manage_message_history='server' by @DouweM in #12313
  • Apply FunctionToolset(requires_approval=True) to Tool instances passed to the toolset by @dsfaccini in #12378
  • Recognize SummarizingCompaction(keep_user_messages=True) copies only by the marker on their text by @DouweM in https://github.com/pydantic/pydant...
Read more

v1.107.10 (2026-11-04)

Choose a tag to compare

@adtyavrdhn adtyavrdhn released this 04 Nov 11:51
7f3b01a

What's Changed

🛡️ Security

  • 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 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 v1.107.10 and v2.66.0 (pydantic-ai, pydantic-ai-slim).
  • GHSA-5fj9-5pjw-vpwc (medium): The cloud-metadata blocklist that URL downloads check did not include Tencent Cloud's instance metadata address 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 v1.107.10 and v2.66.0 (pydantic-ai, pydantic-ai-slim).
  • GHSA-jp94-mmr3-2pvv (medium): With OpenTelemetry instrumentation enabled for embedding models, embedding spans recorded the 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 v1.107.10 and v2.66.0 (pydantic-ai, pydantic-ai-slim). Restrict access to historical traces and rotate credentials they exposed.
  • GHSA-7494-925c-rrxv (medium): For models whose profile requires tool schemas without $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 v1.107.10 and v2.66.0 (pydantic-ai, pydantic-ai-slim).
  • GHSA-r879-rxgj-3g75 (medium): The Vercel AI and AG-UI adapters took time that grows with the square of the number of parts to load a submitted message, on the event loop, so a single request could delay other work in the same process. You are affected if you pass client-submitted conversations to VercelAIAdapter or AGUIAdapter. Fixed in v1.107.10 and v2.66.0 (pydantic-ai, pydantic-ai-slim).
  • GHSA-4cj4-hgwf-4hc4 (medium): When message history was serialized and loaded again, a tool result containing a JSON object in the serialized shape of one of Pydantic AI's file types was rebuilt as that file. A third party who controls data a tool returns could use this to make the application download a URL on its local network, or make the model provider read an 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 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 v1.107.10 and v2.66.0 (pydantic-ai, pydantic-ai-slim).
  • GHSA-5ppx-9jq6-6pmw (low): allowed_domains on the local web-fetch tool 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 v1.107.10 and v2.66.0 (pydantic-ai, pydantic-ai-slim).

Full Changelog: v1.107.9...v1.107.10

v2.65.0 (2026-11-02)

Choose a tag to compare

@dsfaccini dsfaccini released this 03 Nov 01:25
a56a6d6

What's Changed

⚠️ Compatibility Notes

  • Protect the gh-aw canonical session capture from agent tools by @dsfaccini in #10606
  • Add multimodal notes to AskUserAnswer by @dsfaccini in #11664
  • Count BrowserUse model calls toward the host run's usage and UsageLimits by @mpfaffenberger in #11860
  • Name the V2 replacement when code imports a name V2 removed, like OpenAIModel or MCPServerStdio by @DouweM in #12035
  • Fall back to the next model in ModelFallback when a model refuses, and let ContentFilterError in fallback_on match refusals by @DouweM in #12039
  • Raise ModelQuotaExceededError instead of ModelRateLimitError for Z.ai balance and plan-limit 429s by @DouweM in #12042
  • Support the '12h' prompt cache retention on the OpenAI Responses API for GPT-5.6 and later by @DouweM in #12046
  • Send Google a server deadline of at least 20 minutes instead of the client's read timeout, and warn about httpx's 5 s default by @DouweM in #12057
  • Stop Temporal from retrying model errors that can't succeed on retry by default, via non_retryable_model_errors by @DouweM in #12107

🚀 Features

  • Add WebSocket transport to OpenAI Responses by @dsfaccini in #10444
  • Add ConversationStore, which keeps a conversation as a log of messages stored once, and Conversation.revision to catch conflicting saves by @DouweM in #10658
  • Add multimodal notes to AskUserAnswer by @dsfaccini in #11664
  • Add anthropic_web_tools_allowed_callers to set allowed_callers on Anthropic's web tools by @aweis89 in #10489
  • Support the '12h' prompt cache retention on the OpenAI Responses API for GPT-5.6 and later by @DouweM in #12046
  • Send mid-conversation system prompts natively on OpenRouter OpenAI routes and the Claude models that support them by @DouweM in #12056
  • Name the V2 replacement when code imports a name V2 removed, like OpenAIModel or MCPServerStdio by @DouweM in #12035
  • Fall back to the next model in ModelFallback when a model refuses, and let ContentFilterError in fallback_on match refusals by @DouweM in #12039

🐛 Bug Fixes

  • Raise ModelContextWindowExceededError and ModelQuotaExceededError for Bedrock errors on AsyncAnthropicBedrock by @DouweM in #12040
  • Raise ContentFilterError instead of TypeError when Azure filters a streamed OpenAIChatModel prompt by @DouweM in #12041
  • Re-surface nested RepoContext instruction files after compaction, and stop repeating notes still in history by @DouweM in #11704
  • Include the refusal text in ContentFilterError when the provider also reports a finish reason by @DouweM in #12036
  • Raise ModelQuotaExceededError instead of ModelRateLimitError for Z.ai balance and plan-limit 429s by @DouweM in #12042
  • Protect the gh-aw canonical session capture from agent tools by @dsfaccini in #10606
  • Close PlaywrightBrowser's Chromium and driver when an anyio scope cancels a run by @dsfaccini in #12047
  • Ask Coder to match the surrounding code's style instead of applying SOLID by @mpfaffenberger in #12120
  • Return a Coder tool's error to the model once its retries run out instead of ending the run by @mpfaffenberger in #12121
  • Keep *args and **kwargs docstring descriptions on tool parameters by @Aryapatel-007 in #12126
  • Read tool parameter descriptions from Keyword Args and Other Parameters docstring sections by @roydonsequeira in #10035
  • Let a retired aws_lambda agent loop finish its own cleanup at the grace deadline by @dsfaccini in #12140
  • Treat auth='' as unset alongside client in GitHub, Notion, Ordinal, PostHog, Pylon, and Slack by @mpfaffenberger in #11852
  • Count BrowserUse model calls toward the host run's usage and UsageLimits by @mpfaffenberger in #11860
  • Let FileSystem tools emit their events under PrefixTools by @mpfaffenberger in #11867
  • Forward cache='1h' to Claude Haiku 5.5 on Bedrock instead of snapping it to '5m' by @DouweM in #12045
  • Mark <context> elements in user and tool content as not from the application by @DouweM in #12050
  • Stop Temporal from retrying subclasses of non-retryable errors, like ContentFilterError or a UserError subclass by @DouweM in #12105
  • Stop Temporal from retrying model errors that can't succeed on retry by default, via non_retryable_model_errors by @DouweM in #12107
  • Stop sending a history container on Anthropic requests without code execution by @dsfaccini in #11764
  • Send Google a server deadline of at least 20 minutes instead of the client's read timeout, and warn about httpx's 5 s default by @DouweM in #12057
  • Place the OpenRouter Gemini message cache breakpoint before the final message, where it no longer doubles the billed prompt by @DouweM in #12048

📦 Dependencies

  • Remove the Daytona close() workaround and require daytona>=0.226.1 by @mislavivanda in #11977

New Contributors

Full Changelog: v2.64.0...v2.65.0

v2.64.0 (2026-10-30)

Choose a tag to compare

@dsfaccini dsfaccini released this 30 Oct 23:27
cde1d10

What's Changed

⚠️ Compatibility Notes

  • Raise an error status a retrying transport gave up on as ModelHTTPError, so fallback moves on by @DouweM in #11934
  • Send stored response, item, conversation and container IDs back only to the endpoint that created them, via Model.accepts_server_state by @DouweM in #11774
  • Make json-repair an optional dependency, needed only by Coder and RepairToolArguments by @mpfaffenberger in #11768
  • Turn off strict tool schemas for Coder's tools by default, with Coder(strict=True) to opt in by @mpfaffenberger in #11743
  • Surface every nested RepoContext instruction file between a traversed path and the working directory, without repeating the root file by @DouweM in #11687
  • Add output_speech_timestamps to report word timing of realtime speech, on Azure AI Voice Live by @DouweM in #11685
  • Make the ModelFallback capability constructible from agent specs by @DouweM in #11679
  • Add a first_event_timeout model setting so a stream that never starts can fall back by @DouweM in #11582
  • Dispatch OnlineEvaluation evaluators once per durable run instead of on every replay by @DouweM in #11577
  • Show the harness Planning plan in the plan tools' results instead of a per-request reminder copy by @DouweM in #11521
  • Add agent artifact security scanning for Memory and Skills with ArtifactScanner by @mpfaffenberger in #11470
  • Fix AnthropicModel.count_tokens rejecting all-deferred tool requests and undercounting server tools on the Anthropic API by @dsfaccini in #11343
  • Accept plain functions as evaluators in pydantic_evals by @dmontagu in #10292
  • Remove unresolvable RealtimeEvent from pydantic_ai.agent.__all__ by @adtyavrdhn in #10189
  • Add RunContext.load_capability to load on-demand capabilities from code by @adtyavrdhn in #10090
  • Add a ModelFallback capability, a per-attempt prepare_model_request hook, and RetryModelRequest by @DouweM in #9348

🚀 Features

  • Let @agent.system_prompt(dynamic=True) append its changes with on_change='append' to keep the cached prompt by @DouweM in #11445
  • Pair DetectLoops tool results with the response that made the calls by @mpfaffenberger in #11551
  • Add BoxdSandbox: a workspace that runs Coder, Shell, FileSystem and every other harness capability in a boxd machine by @MichielMAnalytics in #11367
  • Add a ModelFallback capability, a per-attempt prepare_model_request hook, and RetryModelRequest by @DouweM in #9348
  • Add Harness Reminder(scope='conversation') for reminders that stay in the message history by @DouweM in #11642
  • Add output_speech_timestamps to report word timing of realtime speech, on Azure AI Voice Live by @DouweM in #11685
  • Cache the schema of ModelRequestContext durable capability operations across agent construction by @DouweM in #11776
  • Make the ModelFallback capability constructible from agent specs by @DouweM in #11679
  • Report word timestamps of ElevenLabs Agents speech with output_speech_timestamps by @DouweM in #11784
  • Record CI provenance on Pydantic Evals experiment spans with OpenTelemetry vcs.* and cicd.* attributes by @dmontagu in #10294
  • Accept plain functions as evaluators in pydantic_evals by @dmontagu in #10292
  • Add logfire.scorer.score_type and logfire.scorer.run_id to online evaluation result events by @dmontagu in #10293
  • Add the OpenTelemetry test.suite.name and test.case.name attributes to Pydantic Evals spans by @dmontagu in #11538
  • Turn off strict tool schemas for Coder's tools by default, with Coder(strict=True) to opt in by @mpfaffenberger in #11743
  • Make json-repair an optional dependency, needed only by Coder and RepairToolArguments by @mpfaffenberger in #11768
  • Teach Coder to read what the task names, check stated requirements, and delegate only when asked by @mpfaffenberger in #11811
  • Let AskUser answerers receive RunContext by @dsfaccini in #11662
  • Add agent artifact security scanning for Memory and Skills with ArtifactScanner by @mpfaffenberger in #11470
  • Add ModelResponse.model_id recording the model the agent sent each request to by @DouweM in #11807
  • Save function evaluators to Dataset files under the function's name by @dmontagu in #10425
  • State how nested instruction files relate in RepoContext instructions when nested_traversal is on by @DouweM in #11698
  • Add a first_event_timeout model setting so a stream that never starts can fall back by @DouweM in #11582
  • Add sticky to ModelFallback to keep a conversation on the fallback model that answered by @DouweM in #11330
  • Add RunContext.load_capability to load on-demand capabilities from code by @adtyavrdhn in #10090

🐛 Bug Fixes

  • Preserve nested SpanTree attributes during JSON round-trips and mark omitted Google embedding and ElevenLabs usage as unmeasured to avoid false zero automatic costs by @dsfaccini in #11912
  • Fall back from Native Output on Bedrock when the installed botocore predates outputConfig by @DouweM in #11694
  • Close the IPythonKernel client's zmq context on shutdown with jupyter-client before 8.7 by @DouweM in #11716
  • Deliver queued realtime messages after a mid-reply reconnect on OpenAI, Azure and xAI by @DouweM in #11678
  • Keep the Gemini Live event stream alive when a resumed session lost a tool call without an id by @DouweM in #11680
  • Search a mixed CodeMode Tool Search corpus with search_tools, so a native deferred tool no longer declares the sandboxed ones by @DouweM in #11638
  • Release a closed realtime session's run state by reference counting by @TejasS1233 in #11240
  • Show the harness Planning plan in the plan tools' results instead of a per-request reminder copy by @DouweM in #11521
  • Dispatch OnlineEvaluation evaluators once per durable run instead of on every replay by @DouweM in #11577
  • Fix AnthropicModel.count_tokens rejecting all-deferred tool requests and undercounting server tools on the Anthropic API by @dsfaccini in #11343
  • Stop advertising CodeExecutionTool on the Bedrock AnthropicModel clients by @DouweM in #11790
  • Turn on the unified cache setting for GPT-5.6 and later with the Azure OpenAI v1 API by @DouweM in #11749
  • Keep the usage and response ID a primed stream received when the consumer leaves it without iterating by @DouweM in #11783
  • Keep Shell's job files in a private temp directory when the workspace isn't writable by @mpfaffenberger in #11809
  • Inline $refs under propertyNames and contains for OpenRouter Google models and custom prefer_inlined_defs transformers by @jukebox in #10859
  • Strip regex patterns Bedrock rejects from strict JSON schemas in BedrockJsonSchemaTransformer by @adtyavrdhn in #11831
  • Remove unresolvable RealtimeEvent from pydantic_ai.agent.__all__ by @adtyavrdhn in #10189
  • Keep finished BackgroundTools results in the history when a run pauses for deferred tools by @wangrat in #11752
  • Let FileSystem edit_file edit non-UTF-8 text files that read_file reads, honoring expected_hash by @mpfaffenberger in #11851
  • Make SubAgents children that share usage run under the parent's usage_limits by @mpfaffenberger in #11853
  • Publish AgentSpec schema fields for `Adviso...
Read more

v2.63.0 (2026-10-27)

Choose a tag to compare

@dsfaccini dsfaccini released this 28 Oct 00:55
e2fbcf6

What's Changed

⚠️ Compatibility Notes

  • Default CodeMode to dynamic_catalog=True, so a revealed tool is appended instead of rewriting the cached prompt by @DouweM in #11358
  • Pass the Gemini Live reconnect replay to GoogleRealtimeConnection's dial as a keyword-only replay argument by @DouweM in #11515
  • Give Harness LLMReminder a default interval=10 instead of generating on every request by @DouweM in #11641
  • Replay signed and encrypted history data to providers with the same canonical_system by @DouweM in #11504
  • Let a deferred CombinedCapability load as one capability, with members keeping their global ordering by @adtyavrdhn in #10321
  • End a GPT-Live reply that delegated at Live's next pause instead of mid-sentence when the backend answers by @DouweM in #11250
  • Say what the <system> tag means and escape it in user and tool text on models without mid-conversation system messages by @DouweM in #11506

🚀 Features

  • Add task_timeout to Dataset.evaluate to bound each task attempt by @dmontagu in #10291
  • Support ag-ui-protocol 1.0 by @adtyavrdhn in #8627
  • Default CodeMode to dynamic_catalog=True, so a revealed tool is appended instead of rewriting the cached prompt by @DouweM in #11358
  • Pass the Gemini Live reconnect replay to GoogleRealtimeConnection's dial as a keyword-only replay argument by @DouweM in #11515
  • Add OpenAI gpt-6.1-terra model support by @dsfaccini in #11393
  • Add an IPython Code Mode (IPythonCodeMode) alongside the Monty-based CodeMode by @mpfaffenberger in #11570
  • Give Harness LLMReminder a default interval=10 instead of generating on every request by @DouweM in #11641
  • Add a DetectLoops capability to interrupt repeated-action loops by @mpfaffenberger in #11471
  • Add manage_message_history to UI adapters so a server-held history isn't sent to the model twice by @DouweM in #11431
  • Keep strict schemas within Anthropic's and OpenAI's complexity limits, falling back to non-strict output where no mode was chosen by @DouweM in #11464
  • Add a public pydantic_ai.framing module with wrap, neutralize_tags and register_tag, so extensions can frame their own tags by @DouweM in #10957
  • Replay signed and encrypted history data to providers with the same canonical_system by @DouweM in #11504
  • Add PostgresSpendStore so harness SpendLimits budgets can be shared through PostgreSQL by @DouweM in #11527

🐛 Bug Fixes

  • Let a deferred CombinedCapability load as one capability, with members keeping their global ordering by @adtyavrdhn in #10321
  • List the modules Monty 1.2-1.4 added in the CodeMode run_code description by @mpfaffenberger in #11459
  • Require mistralai>=3.1.0 and drop the legacy httpx import so MistralModel imports on fresh installs by @adtyavrdhn in #11549
  • Keep the rate-limit category of Azure too_many_requests Responses failures on openai>=3.32.0 by @dsfaccini in #11532
  • Run ToolOutputLimits spill writes in a durable_operation so replays do not write them again by @mpfaffenberger in #11460
  • Record the instructions a before_model_request hook sent when it rewrote the history with a copy of the request by @DouweM in #11501
  • State a deferred capability's instructions on the request where a before_model_request hook loaded it by @DouweM in #11516
  • Escape framing tags consistently in BackgroundTools, RepoContext, TrajectoryJudge and Planning frames by @DouweM in #11530
  • Record a thinking change after an unanswered ModelSettingsDeltaPart instead of replacing it, so UI streams and hook-rebuilt requests keep it by @DouweM in #11514
  • Make typed turns on Gemini gemini-robotics-er realtime models see the image sent just before them by @DouweM in #11484
  • End a GPT-Live WebRTC sideband session cleanly when the browser hangs up by @DouweM in #11435
  • End a GPT-Live reply that delegated at Live's next pause instead of mid-sentence when the backend answers by @DouweM in #11250
  • Withdraw a pending batch request when the run doesn't end with its PendingModelResponse by @DouweM in #11517
  • Send per-message effort changes through AsyncAnthropicBedrockMantle on Claude Opus 5.5 by @DouweM in #11486
  • Enable thinking-block binding recovery on AsyncAnthropicBedrockMantle by @DouweM in #11487
  • Stop offering AdvisorTool on AsyncAnthropicBedrockMantle, which rejects it by @DouweM in #11488
  • Keep the stored CodeMode REPL state when repl_store load fails, instead of saving a fresh REPL over it by @mpfaffenberger in #11483
  • Fix RuntimeError: coroutine ignored GeneratorExit and an escaping KeyboardInterrupt on Agent.iter() teardown by @pydanty in #10258
  • Return a wrap_run or on_run_error recovery result from run_stream() instead of raising AgentRunError by @DouweM in #11571
  • Say what the <system> tag means and escape it in user and tool text on models without mid-conversation system messages by @DouweM in #11506
  • Escape framing tags consistently in the SummarizingCompaction summarizer transcript by @DouweM in #11583
  • Enforce allowed_patterns on the real path when search_files targets a single symlinked file by @mpfaffenberger in #11639
  • Normalize bare None type arguments in Dataset by @pydanty in #11254
  • Stop offering WebSearchTool and WebFetchTool on AsyncAnthropicBedrockMantle, which rejects them by @DouweM in #11489
  • Don't commit an automatic SummarizingCompaction summary that would make the history larger by @DouweM in #11649
  • Clear the caption in stream_transcripts(delta=True) when a transcript is corrected to empty by @DouweM in #11667
  • Evaluate runs with every OnlineEvaluation capability on an agent, not just the outermost by @DouweM in #11663
  • Omit realtime span model_settings when include_content=False by @DouweM in #11668
  • Call AsyncAnthropicBedrockMantle the Bedrock Messages API, not Claude Platform on AWS, in comments and docstrings by @DouweM in #11482
  • Record the provider item ID on the model's realtime SpeechPart.id by @DouweM in #11675

Full Changelog: v2.62.0...v2.63.0

v2.62.0 (2026-10-26)

Choose a tag to compare

@dsfaccini dsfaccini released this 27 Oct 02:38
79685d1

What's Changed

⚠️ Compatibility Notes

  • Change thinking effort mid-conversation via OpenAI configuration_update and Anthropic per-message output_config, recorded as ModelSettingsDeltaPart by @DouweM in #11111
  • Add ModelRequestRejectedError, ModelAuthenticationError and ModelNotFoundError model error categories by @DouweM in #11081
  • State loaded deferred capabilities' instructions in the system voice, append their changes, and show them to before_model_request by @DouweM in #11103
  • Prefix hosted MCP capabilities' tool names with the service by default, configurable with tool_prefix by @DouweM in #10376
  • Store the SummarizingCompaction summary as a CompactionPart instead of a system prompt, and send provider-less CompactionParts to every model by @DouweM in #10573
  • Deliver realtime SystemPromptParts on each provider's operator channel with session.send() by @DouweM in #11379
  • Preserve run identity when replaying Prefect capability operations by @dsfaccini in #11196
  • Append changed MCP server instructions with on_change='append', and give an id-less URL MCPToolset its hostname as id by @DouweM in #10791
  • Record Harness Memory snapshots in history once per changed run, so requests stay append-only by @DouweM in #10903
  • Tell the realtime model a tool call failed on every provider, so it answers instead of going silent by @DouweM in #11086
  • Reserve DBOS handler step positions during tool ctx.emit() by @dsfaccini in #11157
  • Record GPT-Live speech as one SpeechPart instead of splitting it at each delegated tool call by @DouweM in #11187
  • Deliver ctx.emit() events into the realtime session event stream by @DouweM in #11369
  • Make realtime shared settings warn, raise or ignore like standard ModelSettings: forced tool_choice raises on Gemini Live, parallel_tool_calls=False warns by @DouweM in #11066
  • Keep the prompt cache when changing thinking effort mid-conversation on OpenRouter, via its configuration_update system message by @DouweM in #11426
  • Start GraphBuilder broadcast branches in .to() order instead of reversed, which can break replay of durable workflows paused across the upgrade by @DouweM in #11408
  • Cache on_change='append' instruction blocks behind a second cache point after the static instructions by @DouweM in #11351
  • Build Harness SystemReminders on request-scoped system prompts and deprecate cache_ttl by @DouweM in #9712
  • Add CodeMode(monty_sandbox=...), which takes a remote Monty URL or a function that builds the pool, and deprecate monty_sandbox_url by @adtyavrdhn in #11141

🚀 Features

  • Change thinking effort mid-conversation via OpenAI configuration_update and Anthropic per-message output_config, recorded as ModelSettingsDeltaPart by @DouweM in #11111
  • Add ModelRequestRejectedError, ModelAuthenticationError and ModelNotFoundError model error categories by @DouweM in #11081
  • Export conversation_id_from_run_id from pydantic_ai.durable_exec for third-party durability engines by @DouweM in #11338
  • Run tool calls as Temporal child workflows via child_workflow tool metadata in TemporalDurability by @YHallouard in #7329
  • Restrict tool_choice subsets with allowed_tools on OpenAI Chat Completions to preserve the prompt cache by @DouweM in #10396
  • State loaded deferred capabilities' instructions in the system voice, append their changes, and show them to before_model_request by @DouweM in #11103
  • Prefix hosted MCP capabilities' tool names with the service by default, configurable with tool_prefix by @DouweM in #10376
  • Store the SummarizingCompaction summary as a CompactionPart instead of a system prompt, and send provider-less CompactionParts to every model by @DouweM in #10573
  • Add tool_prefix to the MCP capability by @DouweM in #11364
  • Let SubAgents run delegations as tasks the model can background, steer, answer and cancel, behind task_control by @mpfaffenberger in #10893
  • Add Coder(task_control=) to turn on SubAgents task control by @mpfaffenberger in #11007
  • Add StructuredContent user content, sent as its value to decision models by @DouweM in #10529
  • Use the Pydantic AI Gateway's reported cost for RequestUsage.cost by @adtyavrdhn in #7337
  • Let the model request a harness compaction at a natural break in the task with CompactOnRequest by @mpfaffenberger in #11422
  • Deliver realtime SystemPromptParts on each provider's operator channel with session.send() by @DouweM in #11379
  • Add AnthropicModel.compact_messages() for on-demand native compaction by @DouweM in #11432
  • Add provider batch API support with a BatchModelRequests capability and a get_wrapper_model capability hook by @DouweM in #10672

🐛 Bug Fixes

  • Break GraphRun iterator and ToolManager context reference cycles by @dsfaccini in #11158
  • Preserve run identity when replaying Prefect capability operations by @dsfaccini in #11196
  • Stop Agent.run_stream name inference from keeping deps alive until garbage collection by @dsfaccini in #11219
  • Preserve Cohere response IDs in ModelResponse.provider_response_id by @dsfaccini in #11223
  • Recognize deepseek-flash in DeepSeekProvider by @dsfaccini in #11249
  • Close the whole AgentStream event stream when a consumer leaves node.stream() early by @adtyavrdhn in #11009
  • Close inner event streams so a raising listener cancels in-flight tool calls by @Kludex in #8822
  • Fix thinking=False on OpenRouter's anthropic/claude-sonnet-5.5, which can't turn reasoning off by @dsfaccini in #11261
  • Append changed MCP server instructions with on_change='append', and give an id-less URL MCPToolset its hostname as id by @DouweM in #10791
  • Keep a system prompt after the tool result it followed in AG-UI dump_messages() by @DouweM in #11188
  • Cancel a suspended background job only once when a streamed continuation request fails by @DouweM in #11090
  • Keep billed usage when a run stops before a continued response completes by @DouweM in #11068
  • Classify Azure too_many_requests stream errors for FallbackModel by @pydanty in #11253
  • Make google_proactive_audio work on the default Gemini provider with a separate v1alpha Live client by @DouweM in #11067
  • Keep an empty response in place when AG-UI load_messages() reloads a dumped history by @DouweM in #11192
  • Keep openai_previous_response_id chaining when a rewrite-mode instruction changes beside an on_change='append' block by @DouweM in #11331
  • Record Harness Memory snapshots in history once per changed run, so requests stay append-only by @DouweM in #10903
  • Tell the realtime model a tool call failed on every provider, so it answers instead of going silent by @DouweM in #11086
  • Send the lowest effort for thinking=False on OpenRouter routes that make reasoning mandatory by @DouweM in #11347
  • Reserve DBOS handler step positions during tool ctx.emit() by @dsfaccini in #11157
  • Record GPT-Live speech as one SpeechPart instead of splitting it at each delegated tool call by @DouweM in #11187
  • Keep history before a CompactionPart with an empty summary instead of treating it as a compaction boundary by @DouweM in #11368
  • Replay local history into a fresh Gemini Live session when a reconnect has no resu...
Read more