what stackpulse tracks
Vercel AI SDK releases from GitHub
StackPulse watches Vercel AI SDK release notes and keeps the original source link close to every summary.
The AI toolkit for TypeScript StackPulse turns upstream changelogs into scannable summaries with risky changes, deprecations, migration notes, and source links.
what stackpulse tracks
StackPulse watches Vercel AI SDK release notes and keeps the original source link close to every summary.
upgrade risk
Risky changes are separated from normal feature notes so you can scan upgrade impact before changing production dependencies.
migration notes
Migration steps and recommended actions are only shown when the upstream release notes support them.
This patch release updates the `@ai-sdk/gateway` dependency to version 3.0.237. No other changes are described in the release notes.
Users of this release receive the updated `@ai-sdk/gateway` dependency; the notes do not specify further impact.
This patch release makes failed tool-call error text consistent with the model’s message history by default and improves validation of malformed UI messages. The validation changes stop after the first invalid message or part and reduce unnecessary schema checks and issue reporting, mitigating excessive memory and CPU usage.
Applications using `validateUIMessages` or `safeValidateUIMessages`, and applications handling failed tool calls, are affected by the validation and error-text behavior changes.
Model ID autocomplete suggestions are now derived from each provider’s model catalog declarations. This keeps suggestions in sync with catalog updates while preserving existing suggestions, endpoint-specific lists, and support for custom model IDs.
Users of provider model ID autocomplete are affected; custom model IDs and endpoint-specific suggestions remain supported.
This patch adds tool approval configuration and structured timeouts to `WorkflowAgent`. It also fixes warning logging, step-limit warnings, `prepareStep` behavior, and protection of approved tool arguments during approval replay.
Users of `WorkflowAgent` are affected, particularly those using tool approval, per-step preparation, or model streaming and timeout controls.
Adopt the new workflow configuration options if needed; review the updated `prepareStep` behavior when upgrading.
This patch release updates the package’s dependencies. It includes updates to `@ai-sdk/provider-utils` and `@ai-sdk/openai-compatible`; the release notes describe no other changes.
Users of `@ai-sdk/zai` receive updated provider utility and OpenAI-compatible dependencies.
This patch release updates the `@ai-sdk/provider-utils` dependency to version 5.0.81. The release notes do not describe any other changes.
Users of `@ai-sdk/xai` receive the updated `@ai-sdk/provider-utils` dependency.
This patch release updates the `@ai-sdk/harness` dependency to version 1.0.183. The release notes do not describe any other changes.
Users of `@ai-sdk/workflow-harness` receive an updated `@ai-sdk/harness` dependency.
This patch aligns WorkflowAgent behavior more closely with ToolLoopAgent and the text-generation APIs. It adds constructor-level error and abort callbacks, fixes tool message contents and per-step generation overrides, and improves stream usage aggregation.
WorkflowAgent users are affected, particularly those relying on callback ordering, tool message contents, prepareStep overrides, or streamed token usage.
Review WorkflowAgent callback ordering and per-step overrides if your application depends on their behavior.
Remote file URLs used by `experimental_extract` and `experimental_decide` now require an explicit `experimental_download` hook rather than being downloaded silently through global `fetch`. Inline file data remains usable without a hook; the release also updates `@ai-sdk/provider-utils` and `@ai-sdk/gateway`.
Callers using remote file URLs with `experimental_extract` or `experimental_decide` are affected; callers using inline file data are not.
If you use remote file URLs with these APIs, configure an explicit `experimental_download` hook, ideally backed by a DNS-protected transport.
This patch release updates the package’s dependencies. It includes updates to @ai-sdk/openai-compatible, @ai-sdk/provider-utils, and @ai-sdk/provider; the release notes describe no other changes.
Users of @ai-sdk/zai are affected through its updated dependencies.
This patch release updates the `@ai-sdk/provider-utils` and `@ai-sdk/provider` dependencies. The release notes do not describe other changes.
Users of `@ai-sdk/xai` receive updated provider dependencies.
This patch release updates dependencies for @ai-sdk/workflow. The listed dependency updates are ai@7.0.171, @ai-sdk/provider-utils@5.0.80, and @ai-sdk/provider@4.0.38; the release notes describe no other changes.
Users of @ai-sdk/workflow receive the updated AI SDK and provider packages through this release.
This patch improves validation of malformed UI messages to prevent excessive memory and CPU use. It also changes failed tool-call errors sent through UI streams to preserve the text seen by the model by default, adds download transport and cancellation support, and introduces an experimental extraction API and model specification.
Applications using UI message streams may now expose full failed tool-call error text to clients, while applications validating untrusted UI messages or downloading files may benefit from the resource and transport safeguards.
Review failed tool-call error handling for sensitive details and use safe tool error messages or `redactToolErrors: true` where appropriate.
This patch release makes failed tool-call error text consistent with the model’s message history by default, with an option to redact it. It also limits validation work for malformed UI messages to reduce excessive memory and CPU use.
Applications handling failed tool calls or validating untrusted UI messages are affected; malformed-message validation now stops at the first invalid message or part.
The release notes contain only a patch reference to @ai-sdk/harness@1.0.181. They provide no details about the changes, so specific fixes or improvements cannot be determined.
view details view source on github->Fixed workflow telemetry handling when streaming stops on a client-side tool. The final step's telemetry now ends while preserving its step span and model call in the agent trace; dependencies were also updated.
Users inspecting agent traces for workflows that stop streaming on a client-side tool are affected.
This patch improves performance when streaming long arrays by parsing and validating completed elements incrementally. It also prevents queued chat messages and history from being sent through a replacement transport after a later stop or dispose operation.
Users streaming long arrays with `Output.array` or `streamObject`, and applications that stop or dispose chats with pending send-now operations, are affected.
Update to this release to receive the streaming performance and pending-send cancellation fixes.
This patch updates the Vue package's dependency on `ai` to version 6.0.330. The release notes describe no other changes.
Users of `@ai-sdk/vue` receive an updated `ai` dependency.
This patch release updates the `ai` dependency to version 6.0.330. No other changes are described in the release notes.
Users of `@ai-sdk/svelte` are affected by the updated `ai` dependency.
Fixes excessive CPU usage when streaming long arrays with `Output.array` or `streamObject`. Completed elements are now parsed and validated incrementally instead of repeatedly processing the entire accumulated array after each text chunk.
Users streaming long arrays with `Output.array` or `streamObject` are affected by reduced CPU usage.