Read the updates by product, then download from the corresponding GitHub release. Check that release for installation requirements and file details.
Added categorized Skill Hub discovery and curated catalog support, with assistant search, review and installation of reviewed skills. Added resumable skill creation/import and MCP service creation with connection checks. Added browser design snapshots and follow-up measurements within the current task workspace and execution environment.
Fixed access to installed Skills' bundled scripts, templates and reference files in tasks. Kept each task on its selected Skill version and verified resource integrity. [!WARNING] GitHub IPA 未签名，不能直接安装，也不是 App Store 上传包。 The GitHub IPA is unsigned, cannot be installed directly, and is not the App Store upload package.
Improved turn endings, interruptions and playback recovery in voice conversations. Smoothed the central voice animation. Earlier tool activity now loads as you scroll.
Improved playback continuity for online and enhanced voices in voice conversations. Fixed playback progress accounting after audio conversion, which could prevent a reply from finishing. Added a timely retry message when playback cannot start, avoiding repeated stalled attempts.
Fixed voice conversations skipping a spoken reply after the answer completed. Fixed enhanced voice output being cut short while reading numbers and other content, incorrectly reporting that the text was too long. Kept an actionable retry message when the final reply cannot be read.
Fixed voice conversations skipping spoken replies while speech was still being generated, with actionable playback failure states. Improved read-aloud segmentation for long paragraphs and lists, preserving complete sentences and using separate strategies for enhanced voice models and online services. Preserved the original chat or task mode after leaving voice, and improved scoped tool discovery in Chat mode.
Improved false-interruption handling during voice conversations to reduce unintended pauses and cancelled spoken answers. Voice output now lets you choose a playback source before selecting its voice. Configured online speech services can be selected directly, with separate configuration and voice-pack management.
Fixed microphone startup, recognition renewal and playback state handling in voice conversations. Improved repeated use of enhanced speech: recognition models stay loaded after their first foreground use, while stopping recording releases the microphone. Added speed and pitch controls and improved playback after pausing, stopping and interruptions.
Improved enhanced voice loading, continuous playback and stop response. Fixed Volcengine Agent Plan speech playback and added a configurable OpenAI-compatible speech service address. Office printing and PDF export are grouped in the upper-right menu, with automatic previews and an option to include handwriting.
Location support works with existing background features, allowing tasks to request their location for weather and nearby searches while in the background. Location permission is requested on first launch, with location status available in Background execution settings. Fixed first-use location authorization and cancellation races affecting later requests.
GitHub releases are independent of App Store and TestFlight availability. View GitHub releases
Release information updated 2026-11-07 20:50 UTC
