App Quick Start

From installing the desktop app to completing your first task takes about 10 minutes.

Why Use the Desktop App

The Web client is the "remote control" for your workspace, while the desktop app is the true "execution end." Only the app can:

• ✅ Read and write local files (organize directories, generate reports, modify code)
• ✅ Execute Shell commands (npm install, git operations, run scripts)
• ✅ Connect to IM channels (Telegram, DingTalk, Feishu, Teams, and 14+ platforms)
• ✅ Create scheduled tasks (run automatically every day, Cron expressions, notification push)
• ✅ Configure LLM models (multiple API Keys, capability tags)
• ✅ Install MCP tools (extend AI capabilities)
• ✅ Sandboxed execution (run untrusted code in isolation)
• ✅ Manage account memory (view, edit, graph visualization)

Although the Web client has a lighter interface, all "hands-on" capabilities must be provided by at least one app node.

Step 1 · Download and Install

Get the Installer

Platform Support

Step 2 · Launch and Log In

1. Find Workspace (macOS) / Workspace (Windows) in your application list or Launchpad and launch it
2. The login screen appears on first launch
3. Log in with the same enterprise account you use on the Web client (email/phone number/SSO)
4. After a successful login, the app automatically registers as a node to the Gateway

After logging in, you land on the Work conversation page by default. The left-hand menu, in order, is Work / Search / Workflows / Agents, and the bottom holds ⚙️ Settings / 🌐 Language / 🟢 User avatar.

Step 3 · Configure an LLM Model

An Agent needs at least one LLM to work. The desktop app has a unique "model configuration" feature — you need to provide an API Key.

Recommended First Choice: Claude Sonnet 4

1. Click ⚙️ Settings at the bottom left
2. Select Models from the left-hand menu
3. Click + Add Model
4. Choose the provider Anthropic
5. Paste your Anthropic API Key (obtained from https://console.anthropic.com)
6. Choose the model claude-sonnet-4 or claude-3-5-sonnet
7. Save

Model Selection Cheat Sheet

For detailed selection advice and a capability matrix, see Model Configuration and Supported Models List.

Step 4 · Your First Conversation

Return to Work (the first item in the left-hand menu):

1. Below the input box you'll see:
   • Node name: Workspace (your computer name)
   • Model: the model you just configured
   • Working directory: your home directory by default (click to switch)
2. Enter a task (for example, List all files on my desktop)
3. Press Enter to send

The Agent will:

• Automatically call the List Directory tool
• Stream the execution process in real time
• Output the result when finished

Step 5 · Try Real Tool Calls

Try some more complex tasks to get a feel for the Agent Loop:

Simple Task

                      
                      Create a hello.txt file on the desktop for me, with the content "Hello from Workspace"

                    

→ The Agent calls the Write File tool and automatically creates the file

Commands That Require Confirmation (a dialog pops up the first time)

                      
                      Create a demo directory on the desktop and initialize it as a git repository

                    

→ The Agent calls Bash (a P1-level tool; a permission confirmation dialog pops up the first time)
→ After reviewing the command, click "Allow" or "Always Allow"

Multi-Step Task (Experience the Agent Loop)

                      
                      Help me organize my desktop files:
1. Categorize by extension (images / documents / code / other)
2. Create a subdirectory for each category
3. Move the files into the corresponding directory

                    

→ The Agent automatically breaks this into multiple steps: list files → categorize → create directories → move files
→ You can see each step's tool calls in the conversation

Step 6 · Enable Automation Capabilities

By this point you already know how to use the basic capabilities. Next are the app's unique "superpowers":

🤖 Scheduled Tasks (let the Agent run automatically on a schedule)

Settings → Scheduled Tasks → New. See Scheduled Tasks for details.

💬 IM Channels (let your team use AI in group chats)

Settings → Channels → choose Telegram / DingTalk / Feishu, etc., and configure step by step. See Channel Configuration for details.

🧩 MCP Tools (extend more capabilities)

Settings → Tools → MCP Tools section → Add MCP Server. See Tool Management for details.

🧠 Pre-Registered Subagents (configure specialized roles)

Settings → Subagents → New (such as "Code Reviewer" or "Documentation Writer"). See Subagent Management for details.

🛡️ Runtime Security (adjust tool permission levels)

Settings → Runtime Security → adjust the P0/P1/P2 levels for tools such as Bash/Web Fetch. See Runtime Security for details.

Step 7 · Let the AI Remember You

In a conversation, say:

                      
                      Please remember that I prefer to write code in TypeScript, with a code style of 2-space indentation.

                    

The Agent will call the memory_manage tool to write to account memory. The next time you start a new conversation, this preference is automatically injected, and the Agent will naturally respond according to your preference.

In the app → Settings → Memory, you can view, edit, and delete all memories, and it also supports graph visualization to view the connections between knowledge points. See Memory Management for details.

FAQ

Q: How does the Web client know the app is online?

Web client → Space Management → Work menu to view the node list. Your app should appear as an Agent-type node. If you don't see it, check whether the app is logged in and the network is working.

Q: Let my teammates use my app too

Set the app node's scope to enterprise:

1. App → Settings → Runtime Security → Nodes
2. Change the node visibility to Enterprise
3. Members of the same organization can then see and call this node from their own Web client

Note: Setting it to enterprise means your colleagues' tasks may run on your computer, including file read/write and command execution. Please make sure you trust your team members.

Q: Can I run multiple instances of the app?

Running multiple instances under the same account is not recommended — it creates multiple node registrations and may cause routing confusion. If you need multi-device collaboration, use one app per device.

Q: Can I use it offline?

Most features require a connection to the Gateway and the LLM API, so they are unavailable offline. Some local offline capabilities (such as a local LLM) may be supported in the future.

Next Steps

• To get a full overview of app settings → App Settings Overview
• To learn advanced usage → Conversation Interaction, Subagents
• To manage your organization → Space Management Overview