dalao.ai

Cherry Studio

An AI workstation for every model you useOpen source

About Cherry Studio

Cherry Studio started as a desktop app that put every major model in one window. Its makers now call it an AI workstation: beyond chat, its agents can work with files, run commands and drive a browser. It's open source and runs on Windows, macOS and Linux, with iOS and Android apps as well. The Community Edition is free, and by default API keys and conversations stay on the user's own computer.

EditorialUpdated 8 min read

The Cherry Studio interface

What problem it solves

People who bounce between several AI providers every day often keep a row of browser tabs open, one account and one interface per provider, with their history scattered across all of them. Cherry Studio pulls that into one local app. Users add API keys for OpenAI, Anthropic, Gemini or any compatible service, or connect local models running in Ollama or LM Studio, then put the same question to several models at once and read the answers side by side.

The website claims 300+ models from 50+ providers and 1,000+ built-in assistants (as of 2026-10-25). The English docs say 60+ provider templates, so the two don't agree. Two first-party services come built in: the free CherryAI and CherryIN, the official gateway, priced per model (the docs point to a model marketplace with per-model prices, and different token groups carry different multipliers).

It's well past being a chat window. The README now lists agents first. Each agent gets its own instructions, workspace, model and tool permissions. Agents can work with files, run commands, use browser and computer automation tools, pick up extra abilities through Skills and MCP servers, run on a schedule, and reach messaging platforms through Channels. Knowledge bases, document processing, notes, image generation, translation and voice live in the same app.

The pivot to a workstation came with version 2.0 on August 5, 2026, whose release notes are titled "from Chat to Getting Work Done." It redesigned the interface and added multiple windows and split views. Agents gained support for both OpenAI and Anthropic endpoints, can reuse Claude and Codex subscriptions, and now ship with their runtime built in, so there is far less to install. Agents can also read and maintain knowledge bases so they grow like a wiki, and the old code tools became Code Mate, which configures model services for Claude Code, Codex and similar tools in one click. Cherry also says it rebuilt the data layer so the app stays responsive with a long history and large folder imports no longer freeze the interface.

As for who's behind it: the GitHub organization CherryHQ maintains the project, and the project lead is kangfenmao. The repository dates from May 2024.

Cherry Studio is an AI workstation for Windows, macOS, and Linux. Chat with multiple models, work with agents, build knowledge bases, create images, and automate everyday tasks in one place.

Cherry Studio README

Highlights

Many models, one conversation. One prompt can go to several models, with the answers side by side. Responses render Markdown, highlighted code, formulas and Mermaid charts, and conversations can be exported.

Agents. Each agent has its own workspace and tool permissions. Agents can handle files, run commands, automate a browser or the desktop, use MCP servers and Skills, run on a schedule, and connect to messaging apps through Channels. The README is upfront that what works depends on the provider, operating system and setup, and some features need extra dependencies or system permissions.

Knowledge bases and documents. Files, folders and web pages can become a local knowledge base that conversations and agents can draw on. It reads PDFs, Office documents and images, with OCR, and there are Markdown notes plus a file library for attachments and generated files.

Everyday tools. Image generation and editing, translation of text and documents (including quick translation of text selected in other apps), web search, dictation and read-aloud (local models supported), Mini Apps, and an API Gateway that lets other apps use the models configured in Cherry Studio. The workspace supports tabs, detached windows, a Quick Assistant and custom CSS themes; community themes written for v1 need adapting for the v2 interface.

Visible costs and automation. A usage page totals cost, tokens, requests and daily activity. Scheduled tasks run agents on a cron schedule, Channels connect agents to platforms such as Feishu and Telegram, the selection assistant translates, explains or rewrites any highlighted text, and the Quick Assistant is a small chat window that floats over everything. Deleted files, assistants and topics go to an archive first and can be restored.

Local data, plenty of backup options. The privacy policy says API keys and conversations are stored only on the user's device, with no cloud sync service. The data settings docs list the backup choices: a local folder, WebDAV (Nutstore and 123pan are the examples given) or S3-compatible storage such as AWS or Alibaba Cloud OSS. Any of them can run automatically, anywhere from every few minutes to every 24 hours. Backups are complete by default; a lean mode skips images and knowledge base files. Conversations can also be exported to Notion, Obsidian or SiYuan, and history imported from ChatGPT and Claude.

How it compares

Open WebUI and LobeChat take the web route. Open WebUI's repo describes it as an AI interface supporting Ollama and the OpenAI API; it typically runs on a server and lots of people share it through a browser. Cherry Studio is a native app set up on each computer, so one person can get going faster, settings and data stay with that machine, and nobody has to look after a server first.

That's also the tradeoff. Everyone sets up their own keys and models, and there's no shared entry point or permission model for a team. Cherry's answer is the Enterprise Edition, which the README says adds centralized model management, shared knowledge bases, employee access controls and private deployment.

Since 2.0 it has picked up another role. Code Mate configures model services for command-line coding tools like Claude Code and Codex, and agents can reuse those subscriptions. For people already using those tools, it works more like a local console for models and agents.

Who uses it

As of 2026-10-25 the GitHub repo has about 53,000 stars and 5,100 forks, with about 412 contributors who have GitHub accounts (GitHub API, 2026-10-25); the 511 shown on the website also counts committers without linked GitHub accounts. The latest release, v2.1.7, shipped on 2026-10-19. The iOS and Android apps are developed in a separate repository. The community gathers mainly on Telegram, Discord and QQ.

Who it's for

People with keys from several providers who want to manage and compare models in one place on their own machine. Heavy users who want local knowledge bases, document OCR, translation and notes. Anyone curious about agents that handle files and web pages without writing code. And people who care where their data lives and don't want conversations sitting in someone else's cloud.

Not for

Teams that need one shared web entry point with central account management. People with no API keys who don't want to set up local models. And companies wary of the AGPL that plan to modify and redistribute the code.

Getting started

Installers for each system are on the website. The English quick start's steps: open Settings → Model Provider, pick a provider, enter the API key, select Sync models, add a chat model, and enable the provider; the chat view is then ready. The check button tests a model to confirm the setup works. A model only appears in the model pickers once it's been added to the list. Settings changes apply immediately, with no restart.

The provider settings also include the following. One provider can hold several keys, comma-separated, and the key icon next to the field labels, enables or disables each one, and a specific key can be chosen for testing. If a third-party service gives a full URL like https://xxx.com/v1/chat/completions, usually only the root is entered and the app fills in the path. For a non-standard path, a full URL ending in # stops the app from appending anything. The actual model ID is shown under each name; dated and prefixed variants with similar names aren't merged. Advanced API settings let some providers turn on options such as enable_thinking (whether Qwen3 and similar models reason) or service_tier (OpenAI only).

Local models need Ollama or LM Studio set up first, then added as a provider. Without a key, the built-in CherryAI service is free, and prompts pass through Cherry's servers.

Backups are configured under the data settings. Backup files include the API keys, and the docs warn about this. As of 2026-10-25 the latest release is v2.1.7, from October 19, 2026.

Pricing & license

The Community Edition is free under AGPL-3.0. Commercial use is allowed with full AGPL compliance, which means, for example, publishing the source of a modified version offered as a service. A commercial license that waives those obligations is available through bd@cherry-ai.com. The Enterprise Edition adds centralized model management, shared knowledge bases, employee access controls and private deployment; for purchasing, pricing or a demo, the README says to contact bd@cherry-ai.com. Model usage is billed by the user's own provider accounts.

Risks & limits

The privacy policy lists two kinds of data collected by default. Basic operational information covers the software version and aggregated feature usage. Product improvement information covers feature usage activity and frequency, plus error logs and crash reports. Cherry says all of it is anonymous and excludes conversations, API keys and personal information, and both kinds can be switched off under Settings → Data Settings → Privacy Settings.

With the built-in model services, prompts are relayed through Cherry's official servers. Cherry says it doesn't log them and discards them once the request finishes. Backups contain the API keys. Once conversations are connected to Notion, Yuque, Joplin, Obsidian or SiYuan, those services' own privacy policies apply. Privacy questions go to privacy@cherry-ai.com. For users upgraded from 1.x, the data settings still show the V1 legacy data.

Our verdict

Cherry Studio is aimed at individuals who bring their own keys: a local multi-model desktop app whose agents and knowledge bases correspond to the official workstation label. Team use involves a choice between the Enterprise Edition and a self-hosted web app, plus the AGPL terms; the default data collection can be switched off in the privacy settings.

Why it stands out

Cherry Studio is an AI workstation: a desktop app for Windows, macOS and Linux, plus mobile apps for iOS and Android. It puts multi-model chat, tool-using agents, local knowledge bases, translation and MCP extensions in one place, and ships with more than 1,000 preconfigured assistants.

Started by kangfenmao and now built with 511 contributors on GitHub, it earned unusually strong word of mouth outside China for a project that started on Chinese forums, largely because the defaults are sensible and the app respects a user's own API keys.

One app for desktop and mobile brings multi-model chat, agents, local knowledge bases and MCP extensions together, using your own API keys.

Sources

Official

  1. Cherry Studio READMEGitHub · Accessed Oct 25, 2026
  2. Cherry Studio - The All-in-One AI Workstationcherryai.com · Accessed Oct 25, 2026
  3. Quick Start | Cherry Studio Docscherryai.com · Accessed Oct 25, 2026
  4. Model Service Settings | Cherry Studio Docscherryai.com · Accessed Oct 25, 2026
  5. Quick Start | Cherry Studio Docscherryai.com · Accessed Oct 25, 2026
  6. Model Service Configuration | Cherry Studio Docscherryai.com · Accessed Oct 25, 2026
  7. CherryIN | Cherry Studio Docscherryai.com · Accessed Oct 25, 2026
  8. Model Service Settings | Cherry Studio Docscherryai.com · Accessed Oct 25, 2026

Updates & reading