Cherry Studio
Cherry Studio 2.0.0 is a cross-platform AI workspace for cloud and local models, multi-model conversations, assistants, knowledge bases, agents, MCP tools, documents and code tasks.
What Cherry Studio provides
Cherry Studio combines cloud and local model endpoints, conversations, assistants, document knowledge bases, Agent tools, MCP integrations and coding workflows in one desktop workspace. Start with one low-limit provider and a non-sensitive document so each data path can be checked before real material is connected.
Model, knowledge-base and tool boundaries
Cloud models, embeddings, search, image features, WebDAV and MCP services may receive prompts, document fragments or files. A desktop interface and local configuration do not mean every operation stays on the device. Agent and code tools can also reach files, commands or networks, so use a dedicated test directory and human confirmation for write actions.
Release and licensing notes
Version 2.0.0 adds multi-window and split views, an Agent runtime, knowledge-base maintenance and Code Mate across Windows, macOS and Linux builds. The community edition uses AGPL-3.0; enterprise, commercial exceptions and third-party model services have separate terms. Review date: 2026-08-23.
Save to your cloud drive
Open the cloud drive to get the file directly, or save it for convenient access on another device.
Quark Cloud Drive
RecommendedSave Cherry Studio to this cloud drive
Baidu Netdisk
Save Cherry Studio to this cloud drive
Cherry Studio 2.0.0 installation and Agent permissions guide
Install the build for the platform, test one provider and one non-sensitive document, then enable Agent, MCP and coding tools one at a time with limited permissions.
Before you start
- Confirm the Windows, macOS or Linux architecture and reserve space for the application, sessions, indexes and local models.
- Prepare a low-limit API key or local model endpoint and a test folder containing no personal or production data.
- Review AGPL-3.0, the selected provider terms and each MCP or tool service before connecting accounts.
Installation steps
- 01
Choose the matching build
Select the Windows installer or portable build, the correct macOS chip package, or the Linux AppImage, DEB or RPM package, then verify the displayed version.
- 02
Add one test model
Configure one local model or limited API endpoint with its base URL, model name and timeout. Keep the key out of chats, screenshots and shared configuration files.
- 03
Create an isolated knowledge base
Import a small copy of a document, check parsing, citations, deletion and backup, and confirm whether embedding and reranking are local or remote.
Quick start
- 01
Compare providers with sample data
Send the same non-sensitive prompt to one or two providers only when needed, and record which service receives the prompt and attached context.
- 02
Limit Agent and MCP access
Give tools a dedicated test folder and the smallest network scope, begin with read-only tasks and require confirmation before file, command or account actions.
- 03
Verify backup and restore
Back up sessions and configuration, test restoration in a blank profile and protect any backup that contains keys, conversations or knowledge-base indexes.
Usage tips
- Multi-model comparison may send the same prompt to several providers, so use separate test and production contexts.
- Agent, MCP and code tools are executable integrations rather than visual features; review their permissions after every update.
- Community AGPL obligations, enterprise exceptions and provider billing or data policies should be tracked separately.
Troubleshooting and uninstall
Why does a model connection fail or produce different results?
Check the endpoint, model identifier, account quota, proxy, timeout and retry settings. Disable parallel providers and reproduce one short request before rotating an exposed key.
What should I do if an Agent changes the wrong file?
Stop the task, revoke the tool permission, restore a copy or version-control checkpoint, narrow the working directory and disable automatic command approval while reviewing MCP configuration.
- Export data and revoke accessExport required sessions and knowledge bases, revoke API keys, subscriptions, MCP, WebDAV and code-tool access, and protect backups as sensitive files.
- Remove local data deliberatelyQuit Cherry Studio, uninstall it or remove the portable directory, then clear caches, indexes and logs only after confirming that no recovery copy is needed.
Frequently asked questions
Does local installation keep every conversation and document on the device?
No. Cloud models, search, image features, embeddings, Agent tools and MCP services can receive prompts or context. Use local models and local processing where appropriate and review each provider path.
What can an Agent or MCP tool access?
The enabled tool determines the boundary. A tool may read files, run commands or use a network service, so start with a dedicated test folder and least privilege.
Does the download require an extraction code?
The Quark entry does not require one; the four-character code for the Baidu entry is shown beside its download entry.