Skip to content
Agents tracked: 501 Releases (7d): 777 Agent status: no reported issuesSince last limit reset: Claude 10d · Codex 8d For AI agents: Markdown Updated Nov 21, 2026

Cherry Studio changelog: what's new each month

Every stable Cherry Studio release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 12 months covered; the current month updates daily.

November 2026 · so far

4 releases: 2.2.0 → 2.2.3

November brought a major marketplace for skills and resources, improved task management with dedicated scheduling and history views, and significant enhancements to agent capabilities including memory editing and avatar customization. The month focused on streamlining access to extensions and giving users better control over agent behavior and approvals.

Action needed
  • Computer Use is now enabled by default and requires approval for each app accessed; previous behavior can be managed in Settings.
  • Agent system and global package installs now require approval in normal modes and run in Full Access mode instead of being blocked.

Highlights

  • New Marketplace centralizes browsing and installation of skills, agent templates, MCP resources, and external skill sources.
  • Scheduled Tasks now have a dedicated sidebar with full run history, management tools, and conversation access.
  • Agents can now view and edit their memory directly from configuration, and use custom image avatars displayed throughout the app.
  • Computer Use is on by default with per-app approval prompts, letting you control access session by session.
  • Session details now show grouped subagents, task progress, context usage, and Computer Use operations with app icons.

New

  • Marketplace for unified discovery and guided installation of skills, MCP resources, and agent templates.
  • Notification drawer gathers task results and official announcements with unread indicators and multilingual support.
  • Scheduled Tasks sidebar page with paginated run history and management for completed or missed tasks.
  • Pi Durable runtime as default for new agents, providing native session persistence and independent background subagent streaming.
  • Agent memory viewing and editing from agent configuration.
  • Support for uploading image avatars during agent creation, displayed in catalogs, selectors, and conversation headers.
  • Rapid-MLX available as a keyless built-in local model provider.
  • File preview for PPTX slide decks with page navigation, zoom, and thumbnails.

Improved

  • Agent sessions now show grouped subagents, task progress, and context usage in compact cards.
  • Computer Use calls display the applications they accessed, with icons shown in session details.
  • DeepSeek agents now show intent lines above tool calls, with sessions keeping tool steps expanded by default.
  • Agents and assistants can use custom image avatars, with existing emoji avatars unchanged.
  • Workspace and agent groups can be pinned to a top Pinned section with expandable tasks.
  • Conversation grouping and sorting are now independent with options for last updated, creation date, name with natural number order, or manual ordering.
  • Automatic command approval now parses Bash and PowerShell with shared risk rules, showing task context and working directory.
  • File paths and Markdown links with line numbers now jump to and highlight the referenced line in preview.

Fixed

  • CherryIN Claude Sonnet 4.5 tool requests no longer fail with Bedrock parameter validation errors.
  • System and global package installs now work properly in normal modes instead of being blocked.
  • File sizes now use consistent formatting across the app.
  • Session details typography, icon alignment, and spacing polished for better presentation.
  • Browser conversation sidebars now properly support multiple retained tabs with unified styling and context menus.

October 2026

6 releases: 2.1.5 → 2.1.10

October brought major expansions to Cherry Studio's capabilities across voice, mobile connectivity, and agent automation. The month focused on enhancing communication channels, improving file handling, and adding new language and provider support.

Action needed
  • Custom Pi scripts must be updated to use the native codemode API after upgrade to Pi runtime 1.0.
  • DashScope rebranded to Qwen AI Platform with updated endpoints and icons.

Highlights

  • Upload files and images to remote agent sessions from mobile devices with resumable transfers up to 1 GiB.
  • New voice features including Kokoro and Qwen3 text-to-speech, local FunASR transcription, and streaming audio playback.
  • Added WeCom and DingTalk channels with QR code setup for agent conversations and notifications.
  • Expanded provider support with QwenCloud, Tencent Yuanbao Search, Desearch, and SearchApi.
  • Upgraded Pi runtime to version 1.0 with native sandboxed Code Mode and MCP SDK v2 support.

New

  • WeCom and DingTalk channels with QR setup, streaming replies, and attachments.
  • Kokoro Chinese-English TTS with 103 voices and Qwen3-TTS for Apple Silicon.
  • Local FunASR speech transcription on Windows, Linux, and macOS.
  • MCP SDK v2 support with interactive requests, validated forms, and cancellation.
  • System Doctor AI analysis section that reads diagnostics and proposes bounded fixes.
  • Scheduled task creation and management through agent chat with interactive task cards.
  • Italian (it-IT) full UI translation and emoji picker support.
  • File commenting in agent files with line-level hover comments.

Improved

  • Mobile device onboarding with platform-specific downloads and automatic previews.
  • Mermaid diagrams render faster and now match app theme automatically.
  • Agent file pane with tabs, breadcrumbs, tree resizing, and autosave/indent/word-wrap settings.
  • Voice dictation: hold mouse button in text input to start, release to transcribe.
  • Streaming speech synthesis with real-time audio playback.
  • Selection Assistant: hold press to reorder lists; enabled and disabled actions now shown together.
  • Knowledge bases can be dragged between groups with reorderable group columns.
  • Proxy connection testing in General Settings with PAC proxy and direct fallback support.

Fixed

  • Search no longer fails for programming languages with hash symbols like C# and F#.
  • Client-refusal errors now show correct permission guidance instead of misleading invalid API key message.
  • Windows restore failures when browser storage is locked by current session.
  • Local speech recognition no longer loses second half of long utterances.
  • Message list no longer jumps to first message while scrolling conversations with code blocks.
  • Missing or duplicated response text when stream replay overlaps incoming output.

September 2026

8 releases: 2.0.11 → 2.1.4

September brought major new capabilities including device connections for mobile integration, a system doctor for troubleshooting, and built-in browser support for agents. The month also expanded model support significantly with new providers and model additions.

Highlights

  • Agents can now use a built-in browser to navigate pages, find elements, and inspect content while showing their actions with an animated pointer.
  • Device Connections let you pair a mobile client to your desktop and use agents over an encrypted LAN connection.
  • System Doctor provides AI-powered diagnosis and guided fixes for installation, permissions, storage, providers, and other system issues.
  • New office-transform skill lets agents extract and work with content from worksheets, documents, PDFs, and slides.
  • Expanded model support including GPT-6 Astra, Claude Opus 5.5, DeepSeek V4.1, Grok 4.7, and many others with reasoning controls and web search.

New

  • Cherry Cloud sign-in for entitlement-aware agent models with automatic local Gateway startup and quota-aware model selection.
  • Settings now include a sidebar search that finds any setting across all sections with keyboard navigation.
  • Scheduled tasks now support multiple times per day for daily, weekday, and weekly presets.
  • Markdown export for selected messages with multi-select action bar supporting export to Notion, Yuque, Obsidian, Joplin, and Siyuan.
  • Select region feature to pick and quote worksheet ranges, Word paragraphs, PDF pages, or slide shapes as precise structural references.
  • Local ComfyUI image provider for generating images from saved workflow models.
  • System OCR on macOS and Windows to select and copy screenshot text without local model downloads.
  • Pin shortcuts for Apps, Mini Apps, Agents, Assistants, Knowledge Bases, Topics, Agent Sessions, Files, and Code CLI tools in the sidebar.

Improved

  • Translation model now has its own temperature, top-p, and reasoning effort settings editable from the translation panel.
  • Markdown previews outside chat now use the same rich renderer as chat messages.
  • Sent image attachments appear as compact filename chips with real image thumbnails, including after reopening conversations.
  • Agent sessions now label self-started replies to show why a new message appears (Goal round N or Continued after background work).
  • Chat now preserves partial responses and shows a clear, retryable error when a provider stream ends without a finish reason.
  • Skills can be installed from exact supported paths with protected automatic file saving and manual remote updates.
  • Device connections support IPv4/IPv6 over reachable LAN or externally configured VPN with automatic address prioritization.
  • File list now supports Ctrl+A (Cmd+A on macOS) to select all files in the current view.

Fixed

  • Cherry Studio failing to start after an upgrade when the built-in Support Agent still needed to be created.
  • Middle-click autoscroll in the chat message list no longer vibrates or jumps.
  • Tables from PDFs imported through Mistral OCR being dropped; table contents are now indexed as Markdown tables.
  • Pi requests failing with MissingSessionID by preserving the agent session header.
  • Notion export now waits for the page body to be written and reports failed writes as errors.
  • Windows cleanup now tolerates transient DLL release failures after an inference utility process exits.

August 2026

11 releases: 1.9.13 → 2.0.10

August brought Cherry Studio to version 2.0, a major redesign with rebuilt interactions, multi-window support, and integrated agent runtime. The month continued with steady refinements across chat, providers, knowledge bases, and tool handling.

Highlights

  • Version 2.0 release features upgraded visual design, multi-window and split-view layouts, and built-in agent runtime eliminating setup overhead.
  • Support for reading and maintaining knowledge bases that grow like a wiki, plus one-click previews for research, documents, presentations, and code.
  • Code tools upgraded to Code Mate with dynamic parameter display and one-click model configuration for multiple providers.
  • Built-in web search added for DeepSeek, OpenRouter, and Dashscope; Firecrawl integrated as a URL fetch provider.
  • Smooth reasoning effort control in composer and skill global enablement controls for easier agent customization.

New

  • Multi-window and split-view layouts for multitasking between work and reviewing progress.
  • Knowledge base reading and maintenance capabilities for continuous knowledge growth.
  • Built-in agent runtime eliminating dependency installation and environment setup.
  • Web search support for DeepSeek, OpenRouter, and Dashscope with Firecrawl as URL fetch provider.
  • Smooth wheel control for reasoning effort in the composer.
  • Floating task progress capsule for agent operations.
  • Skill global enablement controls for easier feature management.
  • Selective cache and legacy data cleanup options.

Improved

  • Chat interface redesigned for clearer organization of chats, roles, projects, and tasks.
  • Code tools upgraded to Code Mate with dynamic parameter display for each model.
  • Provider support expanded to work with more models and reuse Claude and Codex subscriptions.
  • One-click previews added for research, documents, presentations, data analysis, and coding results.
  • Image generation now dynamically shows available parameters for each model.
  • Agent runtime performance improved with lazy loading and reduced cold idle memory.
  • Message tools, artifact pane, and trace page optimized to skip redundant re-renders.
  • Tab switching improved to eliminate IPC and re-render overhead under Activity keep-alive.

Fixed

  • Prevented renderer out-of-memory errors during long streaming responses.
  • Fixed AI diagnosis misclassifications and improved provider error reporting.
  • Corrected reference context leaks in collapsed preview displays.
  • Fixed stale dashboard recovery across windows.
  • Resolved spell check highlighting and multi-language issues.
  • Repaired literal autolinks that were swallowing emphasis markers in chat markdown.

July 2026

1 release: 1.9.12

July brought refinements to model compatibility and tool interactions, with fixes focused on supporting newer AI models and improving the agent experience.

Highlights

  • Added support for Claude Fable series and Opus 5 models with proper capability detection.
  • Fixed reasoning effort parameter forwarding for DeepSeek V4+ with OpenAI-compatible providers.
  • Improved tool interactions by preventing accidental toggles when approving actions in collapsed groups.
  • Enhanced agent mode to include Cherry Studio identity headers in requests.

Improved

  • Added native Anthropic /v1/models fetcher for better model discovery.
  • Improved SVG image handling to bypass Content Security Policy restrictions.
  • Enhanced file extension handling to preserve leading dots in saved images.

Fixed

  • Fixed DeepSeek V4+ reasoningeffort parameter support for OpenAI-compatible providers.
  • Fixed Claude Fable and Opus 5 model capability detection.
  • Fixed AccordionTrigger toggle behavior when clicking approval buttons in collapsed group headers.
  • Fixed anthropicApiHost configuration for NewAPI provider's Anthropic endpoint.
  • Fixed SVG data URL image downloads blocked by Content Security Policy.
  • Fixed file extension preservation in saveBase64Image and savePastedImage functions.

June 2026

2 releases: 1.9.9 → 1.9.11

June brought model updates and fixes focused on improving compatibility across providers and fixing issues with agent sessions and display rendering.

Highlights

  • Added MiniMax M3 model support with thinking process capabilities.
  • Fixed agent session message cleanup and topic naming settings.
  • Improved Gemini mini-program display and API key handling.
  • Added support for Anthropic Opus 4.8 adaptive thinking.
  • Fixed HTML artifact handling for external browser viewing.

New

  • MiniMax M3 model added to system model registry.
  • MiniMax M3 thinking process support for reasoning tasks.
  • Anthropic Opus 4.8 adaptive thinking support.

Improved

  • Pruned legacy MiniMax models, retaining M2.7 and M3 for cleaner model selection.
  • Improved custom provider model group inference.
  • Enhanced Electron user agent detection for Gemini mini-program rendering.
  • Better OV OCR availability detection for CPU systems.

Fixed

  • Agent session messages now properly cleaned up on deletion.
  • Topic naming settings now respected for agent sessions.
  • Gemini API key encoding in model list requests.
  • Oversized text display issues in Gemini mini-program.
  • HTML artifacts now open correctly in external browsers instead of using external link behavior.

May 2026

4 releases: 1.9.5 → 1.9.8

May brought focused improvements to Cherry Studio's core functionality, with fixes for web search, model provider integrations, and multi-model chat layouts. The month emphasized stability across various AI provider connections and chat features.

Highlights

  • Fixed web search tool integration to prevent empty tool lists in certain configurations.
  • Improved multi-model chat layout to prevent unwanted scrolling in horizontal view.
  • Enhanced reasoning model support for SiliconFlow DeepSeek and Zhipu models.
  • Fixed knowledge base document handling to preserve HTTP URLs correctly.
  • Resolved provider cache issues that now properly refresh when settings change.

New

  • Added timeoutminutes parameter to cron tool for scheduled task control.
  • Support for Grok 4.3 reasoning effort configuration in xAI responses.

Improved

  • Improved Vertex model list fetching and service account setup process.
  • Enhanced GitCode sync reliability for CI workflows.
  • Better multi-turn image editing by converting assistant image blocks to base64.
  • Improved readability of code blocks in light theme.
  • Better OpenClaw external install prompts for clarity.
  • Enhanced tool component styling and functionality.

Fixed

  • Fixed outer scrolling issue in horizontal multi-model layout.
  • Prevented crashes when model provider information is not found.
  • Fixed Anthropic SDK baseURL to strip trailing /v1.
  • Corrected Ctrl key recognition using uniKey value.
  • Fixed thread idle timeout handling in stream chunk adapter.
  • Resolved image block conversion issues for assistant responses in multi-turn conversations.

April 2026

4 releases: 1.9.1 → 1.9.4

April brought major enhancements to Cherry Studio's agent capabilities, introducing a redesigned Skills system, the new CherryClaw autonomous agent, and AI-powered error diagnosis. Multiple model integrations and stability improvements rounded out the month.

Action needed
  • Plugin system renamed to Skills; users must access Skills management through the new unified interface instead of the previous plugin marketplace

Highlights

  • Plugins system renamed to Skills with a unified management interface replacing the marketplace
  • Introduced CherryClaw autonomous agent with personality system, scheduled tasks, and multi-channel integration (Telegram, Discord, Slack, WeChat, Feistel, QQ)
  • Added built-in Cherry Assistant for issue diagnosis, page navigation, FAQ collection, and bug reporting
  • Launched AI-powered error diagnosis that classifies issues and provides step-by-step fix suggestions
  • Added flomo as built-in MCP server and MCPWorld marketplace for expanded integrations

New

  • CherryClaw autonomous agent with personality system and multi-channel support
  • Built-in Cherry Assistant for diagnostics and support tasks
  • AI-powered error diagnosis with classification and fix suggestions
  • Flomo as built-in MCP server for note capture
  • MCPWorld marketplace for MCP server discovery
  • Syntax highlighting and diff viewer in tool output display
  • Topic pinning feature to organize conversation lists
  • Support for agent-authored skills via Skills tool

Improved

  • Tool output display now includes syntax highlighting, diff viewer, line numbers, and file type icons
  • Per-agent skill enablement with workspace symlinks
  • Language selection persists in overview pages
  • Agent settings changes sync to active sessions
  • Model list now includes DeepSeek V4+ with reasoning effort, MiMo V2.5, and Mistral Small 2603
  • File drag-and-drop uploads support case-insensitive extensions when editing messages
  • Provider API keys now rotate when comma-separated in configuration
  • Session instructions preserved in Soul Mode for agents

Fixed

  • Quick panel no longer loses model selection when closing
  • Custom parameters no longer dropped by CherryCIN/NewAPI endpoints
  • Claude Code spawning now correctly resolves sharp image library dependencies
  • Kimi K2.6 model configuration fixed to disable unsupported temperature and top-p parameters
  • Image generation no longer gets stuck with pending placeholders for GPT image models
  • Topic naming no longer retries indefinitely with corrected timeout handling

March 2026

9 releases: 1.7.22 → 1.8.4

March brought significant improvements to agent capabilities, model support, and stability. The month saw the introduction of an independent Agent module, expanded model compatibility, and numerous fixes to core functionality.

Highlights

  • Agent feature separated into an independent module with dedicated page and route for better organization.
  • Added configurable maximum tool calls setting for agents to control execution limits.
  • Knowledge base REST API endpoints now available for programmatic access.
  • Linux support added for Selection Assistant functionality.
  • Expanded model support including MiMo-V2 Pro/Omni, MiniMax M2.7, and improved reasoning model detection.

New

  • GitHub PR review skill (gh-pr-review) for structured PR reviews.
  • Configurable max tool calls setting for agents.
  • Knowledge base REST API endpoints.
  • Selection Assistant now supports Linux.
  • Thought signature persistence for Gemini conversation replay.
  • Clickable file paths in agent tool outputs and inline code.
  • OpenClaw binary download install with auto-update check.
  • Feishu user mentions in issue notification cards.

Improved

  • MCP SDK updated with security vulnerability resolved.
  • Improved Qwen 3.5 reasoning model detection and thinking parameter handling.
  • Enhanced link parsing to handle invalid parenthesized links.
  • PDF processing made more robust for aggregator providers.
  • Default request timeout increased from 10 to 30 minutes.
  • MCP tool parameter parsing and nested schema display corrected.
  • External editor now opens in new window instead of reusing existing one.
  • Directory Select options now render properly with optionRender.

Fixed

  • MCP SDK security vulnerability resolved.
  • Translate button now disabled when sending message is disabled.
  • CodeMirror highlight crash fixed.
  • Message search crash when clicking topic in MessageHeader resolved.
  • Horizontal scroll prevented in agent chat messages container.
  • Windows test failures and code tool launch errors on Windows fixed.

February 2026

3 releases: 1.7.17 → 1.7.19

February brought improvements to code tool support, external editor integration, and better handling of model-specific settings. The month focused on fixing compatibility issues and refining the agent workspace experience.

Highlights

  • Added OpenCode CLI tool support for code tools
  • New ability to open files in external editors from agent workspace
  • Fixed npm detection when launching the app from Dock or Finder
  • Improved reasoning controls for Together and DashScope models
  • Enhanced MCP multimodal tool result handling to prevent API message size issues

New

  • OpenCode CLI tool support for code tools
  • Open-in-external-editor support for agent workspace
  • Auto-i18n workflow now runs daily instead of weekly

Improved

  • Reasoning controls refined for Together and DashScope models
  • Temperature and top-p settings disabled for kimi-k2.5 model to match its requirements
  • MCP multimodal tool results now managed to prevent exceeding API message size limits
  • Database handling improved with proper closure before restore, reset, and factory reset operations

Fixed

  • Uint8Array type incompatibility with BlobPart for tsgo
  • npm detection issue when app launched from Dock or Finder
  • UpdateAppButton display position
  • Circular reference handling in Serializable type migration
  • Code tool test timeout

January 2026

4 releases: 1.7.9 → 1.7.13

January brought expanded model support across multiple providers and improved infrastructure stability. The month focused on adding new AI models, enhancing tool integration, and fixing compatibility issues across platforms.

Highlights

  • Added support for latest Zhipu models and Baichuan models to expand available options.
  • Introduced Anthropic Cache support for improved performance with compatible models.
  • Added Anthropic host support through OpenRouter package integration.
  • Implemented fuzzy search for file lists with relevance scoring to improve navigation.
  • Fixed security vulnerability in DXT plugin system to prevent path traversal attacks.

New

  • Fuzzy search for file lists with relevance scoring.
  • Anthropic Cache support.
  • Anthropic host support via OpenRouter package.
  • Qwen text embeddings support.
  • Latest Zhipu models.
  • Baichuan models.
  • Feishu notification system for workflow failures.
  • TypeScript CLI tool for Feishu notifications with subcommands.

Improved

  • Simplified MCP tool naming format to use mcp{server}{tool} structure.
  • Refactored to use pnpm install for prebuild packages instead of manual downloads.
  • Enhanced macOS modifier key detection in selection service.
  • Improved MCP configuration to support string types for timeout fields.
  • Better reasoning application limited to OpenAI-compatible SDK providers.

Fixed

  • Removed blockmap handling after differentialPackage was disabled.
  • Aligned MCP tool IDs for proper permissions.
  • Fixed HTML content rendering for markdown copy button.
  • Prevented errors on non-Windows systems in OVMS.
  • Fixed path traversal vulnerability in DXT plugin system.
  • Restored patch for claude-agent-sdk compatibility.

December 2025

7 releases: 1.7.2 → 1.7.8

December brought refinements across API handling, model support, and user interface stability. The month focused on fixing edge cases and expanding compatibility with new models and providers.

Highlights

  • Added MCP server log viewer for better debugging and monitoring of tool integrations.
  • Expanded model support to include GPT-5.2 series, Gemini 3 Flash, Qwen3, and Xiaomi MiMo models.
  • Introduced topic batch management mode and merged import/subscribe workflows for easier assistant management.
  • Added tool use mode setting to default assistant configuration.
  • Implemented Knowledge Mem builtin MCP server for enhanced memory capabilities.

New

  • MCP server log viewer for monitoring and debugging.
  • Topic batch management mode for bulk operations.
  • Tool use mode setting in default assistant preferences.
  • Knowledge Mem builtin MCP server integration.
  • Support for GPT-5.2, Gemini 3 Flash/Pro, Qwen3, and Xiaomi MiMo models.
  • Default reasoning effort option to clarify undefined vs none settings.
  • History search now shows keyword-adjacent snippets with highlighted matches.
  • OpenRouter embeddings support in model listing.

Improved

  • Dynamic textarea height adjustment in input bar.
  • Merged import and subscribe popups with export option in assistant management.
  • Budget calculation logic enhancement.
  • Model status lookup performance using Map data structure.
  • Git Bash path auto-discovery and persistence on Windows for Scoop installations.
  • VC++ Redistributable auto-installation without user prompts on Windows.
  • OAuth callback server now bound to localhost for security.
  • Enhanced URL context validation for provider and model compatibility.

Fixed

  • Topic name persisting after deleting last topic.
  • Selected area in code blocks shifting after scrolling.
  • Empty Anthropic-beta headers being sent.
  • Stale Anthropic-beta header for OAuth requests.
  • Wrong type in OpenAI settings caused by type assertion.
  • Global memory settings submission failure.

Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. Cherry Studio: pricing, features and alternatives · All changelogs