Pull Request Explorer

Summary

Fix early stream error handling in OpenAI SDK

Ensures that early error frames from OpenAI streaming responses are treated as retryable API call errors, improving reliability of retry/fallback logic.

Health Assessment

• • PR was merged within 48 minutes, indicating minimal review friction and quick validation of the fix.

Summary

Enforce timeout for streamText step

Fixes a bug where step timers were prematurely cleared, ensuring reliable timeout behavior for text streaming in the v6.0 release.

Health Assessment

• • Quick fix with minimal changes, merged within an hour, indicating low complexity and high confidence.

Summary

Fix React hook preserves chat state when ID undefined

Ensures that passing an undefined chat ID to the useChat hook does not reset the chat instance, maintaining message and stream state across renders.

Health Assessment

• • Quick fix with minimal code changes and no reported issues.

Summary

Fix: include object type in default tool schema

Adds explicit type: object to default tool schema so OpenAI-compatible providers accept tools with empty parameter schemas.

Health Assessment

• • Quick, minimal change merged within an hour, indicating low complexity and high confidence in correctness.

Summary

Fix xAI reasoning extraction in SDK

Ensures the SDK correctly retrieves reasoning text from xAI API responses, preventing silent loss of reasoning data.

Health Assessment

• • Merged within an hour with no rework, indicating a straightforward, low‑risk bug fix.

Summary

Fix signed tool approval metadata preservation

Ensures signed tool approvals are retained across user responses, preventing failures when resuming interactions.

Health Assessment

• • Quick fix merged within an hour, minimal changes, indicating low complexity and high confidence.

Summary

Fix Google provider unsupported Gemini URLs

Ensures Google provider correctly handles Gemini external HTTPS file URLs, preventing unnecessary downloads and improving SDK reliability.

Health Assessment

• • Rapid resolution with minimal rework, indicating a straightforward bug fix.

Summary

Fix MCP callTool hang and leak on abort

Ensures that aborting an in-flight MCP tool call cleans up its response handler, preventing hangs and memory leaks.

Health Assessment

• • Quick fix merged within an hour, minimal code changes, indicating low complexity and high confidence.

Summary

Fix empty assistant message for data-only blocks

Ensures convertToModelMessages does not emit empty assistant messages for data-only blocks, improving message handling and preventing unnecessary empty responses.

Health Assessment

• • Quick fix with minimal changes, merged within an hour, indicating smooth review and low complexity.

Summary

Fix gateway to accept deprecated warnings

This change updates the gateway to handle deprecated warnings in image, speech, transcription, and video responses, preventing parsing errors in v6.0.

Health Assessment

• • Quick, single‑commit fix with minimal code churn and no review delays, indicating a low‑risk, low‑friction change.

Summary

Return validated elements from array output parseCompleteOutput

Fixes a bug where array outputs from generateText returned unvalidated elements, ensuring schema transforms and defaults are applied.

Health Assessment

• • Quick fix with minimal changes, merged within an hour, indicating low complexity and risk.

Summary

Fix LangChain HITL interrupt key mapping

Corrects tool call mapping for LangChain HITL workflows, ensuring proper interrupt handling and preventing orphaned approval IDs.

Health Assessment

• • Rapid resolution within 1 hour, minimal changes, low complexity and high confidence in correctness.

Summary

Fix streamEvents text reset between model invocations

Ensures streaming text is correctly reset for each model call, preventing appended text from previous invocations.

Health Assessment

• • Quick fix merged within an hour, minimal changes, low risk.

Summary

Export isDynamicToolUIPart from ai package

Ensures consumers can import isDynamicToolUIPart, fixing missing export.

Health Assessment

• • Fast merge with no review indicates a straightforward, low‑risk fix.

Summary

Prevent premature tool call finalization in streaming

Fixes a security vulnerability by ensuring tool calls are only finalized after full data is received, preventing incomplete execution. This protects against partial JSON execution and improves reliability.

Health Assessment

• • Fast turnaround with minimal rework; quick merge after single commit.

Summary

Fix Django static assets and vc dev behavior

Ensures Django services bundle static assets correctly and vc dev loads entrypoints when manage.py is nested, improving deployment reliability.

Health Assessment

• • Multiple commits after the first review indicate iterative refinement of Django integration logic.
• • Fast initial review but long cycle time suggests downstream testing or integration challenges.

Summary

Disable immutable assets for export output

Disables immutable asset handling for export and standalone builds to ensure adapters run correctly, preventing runtime errors.

Health Assessment

• • Fast cycle time and minimal rework indicate a straightforward change with quick review.

Summary

Skip directory fsync on Windows to fix persistence

Fixes commit failures on Windows by skipping directory sync, restoring dev cache functionality.

Health Assessment

• • Single small change, quick approval after review, minimal risk.

Summary

Version Packages

Automated release of updated package versions to npm via Changesets action.

Health Assessment

• • Release PR merged quickly with minimal review, indicating an automated release process with low risk.

Summary

Resolve domain projects via project-domains

Improves performance of domain inspection and removal commands by fetching only relevant projects, reducing latency and resource usage for large teams.

Health Assessment

• • Fast review and merge with minimal rework; performance improvement for large teams reduces command latency and resource consumption.

Summary

Deduplicate global flags in CLI suggestions

Prevents duplicate flags in suggested commands, ensuring accurate and conflict‑free CLI usage for developers. This improves reliability of automated suggestions during deployment workflows.

Health Assessment

• • Quick fix with minimal changes and fast review, indicating low complexity and high confidence.

Summary

Clarify BYOK behavior under ZDR filter

Documentation update clarifying BYOK behavior under ZDR filter for the gateway, ensuring consistent wording across releases.

Health Assessment

• • Rapid merge with minimal changes indicates a straightforward documentation update.

Summary

Add example to prepareCall jsdoc

Adds a JSDoc example to the prepareCall method in the ToolLoopAgentSettings type, ensuring documentation consistency across v6.0.

Health Assessment

• • Documentation-only change, minimal review, merged within minutes.

Summary

docs: fix http link to https in deployment guide

Updated the Vercel AI deployment guide to use secure HTTPS links, ensuring accurate and secure documentation for users.

Health Assessment

• • The PR was merged within 0.2 hours, indicating a straightforward documentation change with minimal review friction.

Summary

Fix middleware and message ID examples

Corrects documentation examples to reflect actual API behavior, reducing developer confusion and improving onboarding.

Health Assessment

• • Rapid merge indicates a straightforward documentation fix with minimal review.

Summary

Add MongoDB memory provider to agents memory page

Documentation update adding MongoDB memory provider to agents memory page, enabling developers to integrate MongoDB memory into AI agents.

Health Assessment

• • Quick merge with minimal changes indicates a straightforward documentation update.

Summary

Add model mappings to BYOK provider options

Adds documentation for model mappings in BYOK provider options, ensuring users understand available AI Gateway service features.

Health Assessment

• • Quick merge with minimal changes indicates straightforward documentation update.

Summary

Enhance tool.toModelOutput description in docs

Improves clarity of the toModelOutput API in the AI SDK, aiding developers in correctly using the function.

Health Assessment

• • Quick documentation tweak merged within minutes, indicating minimal impact and low risk.

Summary

Harmonize SAP AI Core provider documentation

Updated documentation to reflect current SAP AI Core provider package name and description, ensuring consistency with AI SDK 5.x/6.x.

Health Assessment

• • Quick merge with minimal changes, indicating straightforward documentation update.

Summary

Show list help for domains ls command

The CLI previously displayed transfer‑in help for `domains ls --help`. This change corrects the output to show the list help, improving user experience.

Health Assessment

• • Quick fix with minimal changes, fast review, no blockers.

Summary

Abort superseded Server Component HMR requests

Improves developer experience by preventing duplicate HMR requests and console errors during rapid edits, reducing build times and debugging noise.

Health Assessment

• • Long review time and large changes indicate significant complexity; the single commit resolved issues without further rework, suggesting a focused fix.

Summary

Add stats comment with partial stats

Ensures pull‑request comments always include bundler statistics even when some jobs are cancelled, improving visibility for reviewers.

Health Assessment

• • The PR had a long review and merge cycle with a large commit after a long gap, indicating slow feedback and potential blocker complexity.

Summary

Enable new scroll handler by default

This change fixes a bug by enabling the new scroll handler, aligning soft-navigation behavior with browser defaults and reducing focus issues.

Health Assessment

• • The review cycle was almost five days long for a small change set, indicating a slow review process but low technical risk.

Summary

Wrap invalid tool input in object

Ensures Anthropic API calls never receive malformed JSON, preventing runtime errors when replaying messages.

Health Assessment

• • Long review cycle (≈4 days) likely due to coordination with external provider changes, but the change is small and low risk.

Summary

Add framework detection during Vercel build

Enables optional, non‑blocking framework detection to improve build accuracy and surface mismatches, allowing teams to roll out or disable the feature via environment variables.

Health Assessment

• • Fast review and single commit indicate low friction; the feature adds ~910 lines across 10 files, a large scope but within a single PR; risk is moderate due to new runtime logic but guarded by opt‑in flags.

Summary

Remove VERCEL_RESOLVE_ROOT_DIRECTORY guard

Enables default monorepo subdirectory builds, simplifying CI for projects in nested directories.

Health Assessment

• • The PR had a long cycle time of over 74 hours with no explicit review, indicating potential bottlenecks. The large code change (~885 lines) increases risk of regressions.

Summary

Upgrade React to latest experimental release

This PR updates the React dependency to the latest experimental version, ensuring compatibility with new features and bug fixes.

Health Assessment

• • Rapid merge by automated release bot with minimal review; small code churn and no complex changes.

Summary

Backport: Update gateway model settings

Automated update of gateway model settings to surface new models in IDE autocomplete.

Health Assessment

• • Quick automated backport with minimal changes and fast review, indicating low complexity and high confidence in the update.

Summary

Fix dev overlay grid spacing for copy button

Adjusts dev overlay grid spacing to prevent copy prompt buttons from overlapping cards, improving UI clarity.

Health Assessment

• • Quick, single‑commit fix with prompt review and minimal scope, indicating low risk and efficient resolution.

Summary

Fix false-negative regex checks with agentic LLM judge

Adjust evaluation criteria to use an agentic LLM judge for specific regex checks, improving accuracy of evals and reducing false rejections of correct solutions.

Health Assessment

• • Long review cycle (88.5h) and single commit suggest slow feedback but minimal iteration; moderate scope of 158 lines across 9 files; no major risk identified.

Summary

Add support for string span IDs in trace backend

The trace viewer now supports CLI-generated traces that use UUID span IDs. This allows developers to debug Vercel CLI diagnostics more effectively.

Health Assessment

• • Quick turnaround with minimal changes; single review cycle indicates straightforward implementation.

Summary

Emit JSON error for unresolved list projects

Provides structured JSON error for non‑interactive Vercel CLI list failures, enabling agents to detect and recover. Suggests actionable commands to locate or link projects.

Health Assessment

• • Fast cycle time and single commit indicate a smooth review process; the new structured error output enhances automation and error handling for non‑interactive CLI usage.

Summary

Wrap invalid Bedrock tool input in object

Fixes Bedrock API errors caused by malformed tool calls, improving reliability for persisted conversations.

Health Assessment

• • The PR required multiple iterations and a significant number of commits to resolve the Bedrock tool call issue, indicating moderate review friction and a large scope for a bug fix.

Summary

Backport support for Alibaba wan2.7 video models

Adds support for Alibaba's wan2.7 text‑to‑video and reference‑to‑video models to the v6.0 release, enabling new video generation capabilities for users.

Health Assessment

• • Backport resolved conflicts manually and was approved after 87 h, with no further changes, indicating a straightforward merge of a new feature into the release branch.

Summary

Update GitHub Actions dependencies

Bumps four GitHub Actions to latest versions, ensuring CI workflows use up-to-date tooling.

Health Assessment

• • Long cycle time due to automated dependency update; minimal code changes; quick approval after review.

Summary

Backport MLflow video embed fix

Fixes video embedding in MLflow documentation, improving clarity for users.

Health Assessment

• • Small documentation change with minimal review; long cycle time likely due to scheduling rather than technical complexity.

Summary

Fix prototype pollution risk in provider JSON parsing

Replaces insecure JSON.parse calls in provider code with a secure parser, preventing prototype pollution attacks and ensuring safer handling of LLM responses.

Health Assessment

• • Rapid review and minimal changes indicate a straightforward, low‑risk fix with no significant rework or blockers.

Summary

Fix extractJsonMiddleware whitespace handling

Ensures streamed JSON responses preserve leading whitespace, preventing incorrect text concatenation in client streams.

Health Assessment

• • Fast review and merge, minimal changes, straightforward bug fix.

Summary

Normalize Anthropic API base URL

Fixes issue with bare Anthropic API URLs missing /v1 path, ensuring correct API calls.

Health Assessment

• • Quick review and merge, low complexity changes with clear testing and description.

Summary

Fix embedMany batch size for Google Gemini

Corrects oversized embedding batches to comply with Gemini API limit, preventing 400 errors and ensuring reliable embedding functionality.

Health Assessment

• • Rapid review and merge with minimal changes; no rework required.