Cherry Studio v2.1.8: File editing, voice features

1. Separate file editing drafts and improved previews

Editing drafts for notes and generated artifacts are kept separate, and failed edits are retained for retries. According to the company, text editing and previews support up to 8 MiB.

RTF, a rich text document format, can be previewed with its formatting intact. Reading position is preserved after refreshing a file, and line spacing in Word documents has been corrected.

The update also fixes content being overwritten when switching notes and editing stopping after dismissing a disk change notification.

2. Voice input and playback as speech is generated

In a text input field, users can hold the mouse button, speak, and release it to transcribe their speech. Users can also customize the voice input shortcut.

Local speech recognition, which converts speech to text on the device, now includes transcription using Qwen3 1.7B.

Speech synthesis now streams generated audio in chunks for immediate playback. Read-aloud now includes code blocks and inline code. Input gains more transcription options, while output improves listening to ongoing speech and replies containing code.

Windows speech recognition now accepts standard 16 kHz WAV input. Added diagnostic information for failed voice tasks and stream timeouts improves troubleshooting alongside input and playback features.

3. Expanded search, scheduled tasks, and remote connections

Tencent Yuanbao Search and DeepSeek Search & Fetch have joined the list of web search providers, expanding search options.

Link fetching checks for Querit and Crawl4AI have also been fixed. Alongside new search providers, the update improves how links are retrieved through existing providers.

Scheduled tasks can be created in conversations with agents and managed through task cards in the chat window. Users can schedule tasks discussed in a conversation and manage them later in the same chat view.

In Feishu group chats, the bot can be configured to respond only when a user @mentions it.

Remote connection invitations support a custom hostname, a name or IP address identifying a device on the network, and a port. This lets users tailor the connection address to their environment.

4. Keep work running, with confirmation before quitting

Closing the last tab opens the launchpad start screen instead of quitting the app. Separating tab closure from app exit lets users move on to their next task after closing the last conversation.

A new feature keeps active AI and background tasks from sleeping. The app now asks for confirmation before quitting with tasks running. Beyond editing and chat, this changes how users decide when to end their work.

Selecting text on screen automatically copies it and displays a quick quote toolbar. This reduces the steps needed to reuse conversation content in another input.

Electron, the framework powering the app, has been upgraded to 44.9.3 for a certificate security fix. This security fix is included in the release separately from new features.

5. Distribution files by operating system

Packages are available for Windows, macOS, and Linux on x64 and arm64. The table below selects representative package formats from the same release and compares arm64 download sizes alongside x64.

Table: v2.1.8 distribution package sizes

Conditions: File sizes displayed on GitHub for the same release

Unit: MB

Files vary by OS, processor architecture, and package format. Windows offers installer and portable versions; macOS offers Apple and Intel chip builds. All three listed formats are smaller on arm64 than x64. Sizes do not indicate installed size or performance.

Version 2.1.8 reduces interruptions between editing retries and subsequent work by separating drafts, retaining failed edits, and opening the launchpad when the last tab closes.