The 41 questions people ask most about YggLab, the AI agent desktop client. Every answer opens with the conclusion, so you can find what you need quickly.
YggLab is an AI agent desktop client that runs on your own computer. It puts Claude Code, Codex, Gemini CLI and 20+ other coding and office agents on one desktop — assign work to one, or have several form a team and finish a job together. It runs on Windows and macOS, all conversations and files stay local, and you can command it remotely from seven channels: Slack, Discord, Telegram, WeChat, WeCom, Feishu and DingTalk.
Build slide decks, clean up spreadsheets, write reports, do research, change code, retouch images, sort out a messy folder. Think of it as a shared desktop where AI handles your daily work — one app instead of a pile of tools and browser tabs. The output lands in your own folders, ready to edit.
Yes. Assign work from your phone through Slack, Discord, Telegram, WeChat, WeCom, Feishu or DingTalk; the task runs on the computer back home and the result comes back to the same chat. You can also schedule tasks — a sales report every Monday at 8 AM, say — and have the result pushed to whichever chat app you picked. Set it once and forget it.
No. YggLab runs entirely on your own computer; conversations, files and output never pass through our servers. You choose which AI to use — Claude, ChatGPT, Gemini, DeepSeek and others — and your API key is stored locally and used to talk to the vendor directly. We aren't in the middle.
Absolutely. YggLab isn't only a coding tool — 20+ built-in assistants cover slides, spreadsheets, documents, academic writing, UI design and data dashboards. Say what you want the way you'd say it to a colleague. No prompt engineering, no command line.
Yes. Sign in to a YggLab account and cloud models are configured for you, so you can start working right after installing without registering with any model vendor. Already have keys? Switch to your own at any time — both paths work.
Cursor and GitHub Copilot are coding assistants inside an editor, built for programmers writing code. Claude Desktop is Anthropic's own client and only runs Anthropic's models. YggLab is a desktop-wide AI workspace: not confined to an editor, it handles decks, spreadsheets, documents and design work; it connects to Claude, ChatGPT, Gemini and 28 model platforms at once; and it lets several agents form a team, question each other and cross-check the work end to end.
No. On startup YggLab finds the Claude Code, Codex and Gemini CLI installs already on your machine and reuses them — nothing reinstalled, nothing duplicated, none of your existing configuration overwritten. MCP tools you set up in another CLI can be scanned and imported in one click.
28 major model platforms are one click away, including OpenAI (GPT), Anthropic (Claude), Google (Gemini), DeepSeek, Alibaba Qwen, Moonshot Kimi, Zhipu GLM and local models through Ollama. On top of that you can point at any OpenAI- or Anthropic-compatible endpoint, so the real number is higher. Mix them on the same desktop: different agents on different models, switchable at any time, locked to no one.
An ordinary work laptop is fine. YggLab itself is light; inference happens in a cloud API or in your own local model, so it doesn't need a GPU. Requirements: Windows 10 or later (x64 and ARM64 installers) and macOS 12 or later (Apple silicon and Intel installers).
Not yet. Right now YggLab ships installers for Windows and macOS only; a Linux build hasn't been released. If you need one, tell us through the feedback form inside the app.
YggLab bills by actual API token usage — you pay for what you use. You can also bring your own API key and talk to the model vendor directly, in which case that usage is settled with them. macOS and Windows are both supported.
Use the Feedback entry inside the app: a title, a description, and screenshots if you have them. It reaches us right away.
YggLab is built by the YGG Lab team in China, and the interface ships in nine languages: English, Simplified and Traditional Chinese, Japanese, Korean, Turkish, Russian, Ukrainian and Portuguese. Alongside OpenAI, Anthropic and Google models it also supports Chinese models such as DeepSeek, Alibaba Qwen, Moonshot Kimi and Zhipu GLM.
Dify and Coze are cloud platforms for building AI apps and workflows in a browser; what you end up with is an application you publish for other people to use. YggLab is a desktop client on your own machine; what you end up with is actual files on your disk — decks, spreadsheets, documents, code. They answer “how do I build an AI product”; YggLab answers “who is doing today's work”.
It's a place to run and coordinate AI agents. YggLab isn't an agent itself — it's where Claude Code, Codex, Gemini CLI and 20+ others live on one desktop, either taking assignments from you or working with each other.
Three kinds of people. Anyone whose week is full of decks, reports and documents. Developers already using Claude Code or Codex who want them working alongside agents from other vendors on one desktop. And anyone who wants their computer to keep working while they're away. No programming needed — describe what you want in plain language.
Go to the download section at ygglab.ygg.com.cn and pick the installer for your machine: Apple silicon or Intel on macOS, x64 or ARM64 on Windows. The page detects your platform and highlights the right one. Download it, double-click to install, and a three-step setup wizard runs on first launch.
Yes, and the Apple silicon build is native (YggLab-mac-arm64.dmg) rather than running through Rosetta. Intel Macs use a separate installer (mac-x64.dmg). macOS 12 or later is required.
Yes. There are native x64 and ARM64 installers for Windows; ARM machines such as Snapdragon X series laptops use the ARM64 build. Windows 10 or later is required.
YggLab updates itself: when a new version is out, download and install it from inside the app rather than coming back to the website. You can also choose whether to receive pre-release builds.
Yes, because inference goes through a model provider's API. If you run local models with Ollama, that inference never leaves your machine. Either way, your conversations and files never pass through YGG Lab's servers.
Yes. DeepSeek is one of the 28 built-in model platforms — select it in settings and add your API key. You can also give different agents different models, one on DeepSeek and another on Claude, say.
Yes. Alibaba Qwen, Moonshot Kimi and Zhipu GLM are all among the 28 built-in platforms, configurable in one click. Beyond that, YggLab accepts any OpenAI- or Anthropic-compatible endpoint, so models that aren't on the list can usually be connected anyway.
Yes. YggLab works with Ollama, so open models running on your own machine can be plugged straight in. Inference then never touches the network, which suits sensitive material.
If you use your own API key, model charges are settled directly with the vendor — YggLab adds no markup and doesn't relay the traffic. If you use cloud models through a YggLab account, usage is deducted from your account balance, and every charge is itemised.
Token usage is built in: today's total, the last 7 days, which model is costing the most, and a line-by-line record — all inside the app.
YggLab tells you and switches to another working model rather than stalling halfway through a task. You can also add several keys and have it rotate between them, and test in one click which ones have expired.
Yes. Several presentation assistants are built in — Slides, Pitch Deck, Morph and 3D Morph — and they generate real .pptx files. You can edit the result directly in the output panel, or hand it off to Keynote or PowerPoint in one click.
Yes. The Spreadsheet, Data Dashboard and Financial Modeling assistants generate real .xlsx files, including KPI cards, charts and three-statement models. The result is editable right in the output panel.
Create a scheduled task and set the time (Mondays at 8 AM, say), the project folder and the model. Results can be pushed automatically to Slack, Discord, Telegram or one of the other channels. A task can also keep the computer awake so it doesn't get skipped because the machine went to sleep.
Message it on any of the seven channels — Slack, Discord, Telegram, WeChat, WeCom, Feishu or DingTalk — and the task runs on the computer back home. You can also scan a QR code to sign into the built-in WebUI from your phone. Files can be sent and received in the chat itself, with large ones split automatically.
Several agents can work in parallel, and separate conversations and scheduled tasks run at the same time. The system tray shows what's currently running, and you can pause everything in one click.
Yes. YggLab finds and reuses your existing Claude Code install — nothing reinstalled, no configuration overwritten, and your current workflow is unchanged. What you gain: Claude Code, Codex and Gemini CLI running side by side and cross-checking each other, models mixed across 28 platforms, in-app editing of decks, spreadsheets and documents, scheduled tasks, seven remote channels, and one-click import of MCP tools you configured in another CLI.
Yes. Add an MCP server once and every agent gets it — no per-agent setup. MCP tools you already configured in another CLI can be scanned and imported in one click, and services that require OAuth are supported too.
Nine: English, Simplified and Traditional Chinese, Japanese, Korean, Turkish, Russian, Ukrainian and Portuguese. There are also five colour schemes, automatic light/dark switching that follows the system, and support for your own CSS themes.
Not unless you've opened permissions that far. There are three tiers: plan only, auto-accept edits, and fully autonomous. By default every tool call needs your confirmation, and you can approve it just once or from then on. Every change is listed in the change panel, can be discarded one by one, and important points can be snapshotted and rolled back.
No. YggLab runs entirely locally; project files and conversation history stay on your computer and never pass through YGG Lab's servers. Only the content you actually send to a model goes out, and it goes straight to that vendor's API by the route you chose.
Yes. YggLab already runs on your own machine, and conversations and project files never pass through YGG Lab's servers. At the model layer you can point at any OpenAI- or Anthropic-compatible endpoint, including a model service your company hosts internally; local models through Ollama work too, so requests never leave the network. On top of that we offer a private integration service for companies, fitted to your model deployment, network and approval process — leave your requirements and contact details on the feedback page.
Share links are temporary — they stop working when the app restarts, and the interface warns you about the risk. Remote channels have a pairing step as well: when someone messages the bot you have to approve them first, and access can be revoked at any time.
Use the Feedback entry inside the app — a title, a description, and screenshots if you have them. You can also go straight to ygglab.ygg.com.cn/feedback.html.
Have a look at the full feature list and Compare, or just install it and see — there are native builds for all four platforms.
Download YggLab