Issues

New issue

How AI SDK maintenance and community facilitation is changing
#17243 · gr2m opened on Jul 14, 2026

AI Gateway: xAI grok-stt speaker diarization (diarize: true) is dropped; no speaker in segments, providerMetadata or raw body
Bug
vercel/ai#22044 · opened on Oct 5, 2026
2
Add 'max' to the portable reasoning level: distinct from 'xhigh' on GPT-5.6, Claude, DeepSeek V4, Kimi K3
Feature
vercel/ai#22043 · opened on Oct 5, 2026
3
harness-pi: host Pi packages are installed on every session despite noExtensions, and MCP sessions fail under Node
Bug
vercel/ai#22040 · opened on Oct 5, 2026
2
Tools registry: update Pushary entry links (AI SDK page, UTM tags)
Documentation
vercel/ai#22031 · opened on Oct 4, 2026
5
bug(provider-utils): parseJsonLines unbounded buffer causes uncatchable OOM (SIGABRT)
Bug
vercel/ai#22028 · opened on Oct 4, 2026
3
WorkflowAgent approval resume fails after intervening user media, after an approved tool already executed
Bug
vercel/ai#22013 · opened on Oct 4, 2026
3
WorkflowAgent duplicates an existing execution-denied tool result when resuming approval
Bug
vercel/ai#22012 · opened on Oct 4, 2026
3
WorkflowAgent omits a length-finished response from returned conversation messages
Bug
vercel/ai#22011 · opened on Oct 4, 2026
3
WorkflowAgent drops assistant text provider metadata from continuation messages
Bug
vercel/ai#22010 · opened on Oct 4, 2026
3
docs: add GenMagic to the OpenAI-compatible providers section
Documentation
vercel/ai#22009 · opened on Oct 4, 2026
3
@ai-sdk/anthropic: the replayed server-side fallback block is only accepted with the server-side-fallback beta, which is added only when fallbacks is set — 400 on the rescue model and on any chain-less request
Bug
vercel/ai#22005 · opened on Oct 3, 2026
3
@ai-sdk/typesafe-ai: read TYPESAFE_AI_BASE_URL like OPENAI_BASE_URL
Feature
vercel/ai#22002 · opened on Oct 3, 2026
4
Memory provider: Mnemoverse (@mnemoverse/ai-sdk), long-term memory shared across tools
Documentation
vercel/ai#22001 · opened on Oct 3, 2026
3
Global AI_SDK_* declarations conflict (TS2403) when two major versions of ai are installed
Bug
vercel/ai#21999 · opened on Oct 3, 2026
2
[anthropic] Structured output through a non-forced answer tool, for models that reject forced tool use
Feature
vercel/ai#21992 · opened on Oct 3, 2026
3
Replayed messages place a deferred provider tool result next to its call, which breaks prompt caching
Bug
vercel/ai#21919 · opened on Oct 1, 2026
3
realtime: no way to redirect or disable the session's audio playback
Feature
vercel/ai#21911 · opened on Oct 1, 2026
3
Gateway: client tool call before gateway-executed tool results loses its tool_result (Anthropic + Google Vertex reject the next call)
Bug
vercel/ai#21895 · opened on Oct 1, 2026
2
AI Gateway: Anthropic system-message placement breaks Codex continuation and compaction
Bug
vercel/ai#21885 · opened on Oct 1, 2026
5
harness-claude-code: a turn after a background task ends empty, because runTurn stops on Claude Code's own task-notification result
Bug
vercel/ai#21864 · opened on Sep 30, 2026
3
Improve error handling API design for composed UI message streams
Feature
vercel/ai#21814 · opened on Sep 29, 2026
Deleted tools send full persisted output to the model after losing toModelOutput
Bug
vercel/ai#21777 · opened on Sep 29, 2026
3
Forward Claude tool allow and deny lists through createClaudeCode
Feature
vercel/ai#21709 · opened on Sep 29, 2026
1
Forward Claude native settings policies through createClaudeCode
Feature
vercel/ai#21708 · opened on Sep 29, 2026
2
Make the AskUserQuestion hook timeout configurable
Feature
vercel/ai#21707 · opened on Sep 29, 2026
1

Navigation Menu

Issues

Issues

AI Gateway: xAI grok-stt speaker diarization (`diarize: true`) is dropped; no speaker in segments, providerMetadata or raw body

Add 'max' to the portable reasoning level: distinct from 'xhigh' on GPT-5.6, Claude, DeepSeek V4, Kimi K3

harness-pi: host Pi packages are installed on every session despite noExtensions, and MCP sessions fail under Node

Tools registry: update Pushary entry links (AI SDK page, UTM tags)

bug(provider-utils): parseJsonLines unbounded buffer causes uncatchable OOM (SIGABRT)

WorkflowAgent approval resume fails after intervening user media, after an approved tool already executed

WorkflowAgent duplicates an existing execution-denied tool result when resuming approval

WorkflowAgent omits a length-finished response from returned conversation messages

WorkflowAgent drops assistant text provider metadata from continuation messages

docs: add GenMagic to the OpenAI-compatible providers section

@ai-sdk/anthropic: the replayed server-side `fallback` block is only accepted with the server-side-fallback beta, which is added only when `fallbacks` is set — 400 on the rescue model and on any chain-less request

@ai-sdk/typesafe-ai: read TYPESAFE_AI_BASE_URL like OPENAI_BASE_URL

Memory provider: Mnemoverse (@mnemoverse/ai-sdk), long-term memory shared across tools

Global AI_SDK_* declarations conflict (TS2403) when two major versions of ai are installed

[anthropic] Structured output through a non-forced answer tool, for models that reject forced tool use

Replayed messages place a deferred provider tool result next to its call, which breaks prompt caching

realtime: no way to redirect or disable the session's audio playback

Gateway: client tool call before gateway-executed tool results loses its tool_result (Anthropic + Google Vertex reject the next call)

AI Gateway: Anthropic system-message placement breaks Codex continuation and compaction

harness-claude-code: a turn after a background task ends empty, because `runTurn` stops on Claude Code's own task-notification `result`

Improve error handling API design for composed UI message streams

Deleted tools send full persisted output to the model after losing toModelOutput

Forward Claude tool allow and deny lists through createClaudeCode

Forward Claude native settings policies through createClaudeCode

Make the AskUserQuestion hook timeout configurable