logo
Development
OverviewDevSpaceWorkSpaceAPI ReferenceBest PracticeChangelog
Search
English
WorkSpace / Work Area / Work / Overview
Work Conversation Overview
Last updated：30 days ago

Work Conversation Overview

What Is Work

Work is the core conversation interface of the workspace — the primary place where you collaborate with the AI Agent, and the module that carries the workspace's "AI avatar" capability.

Unlike a traditional chat window, the Agent in Work has an autonomous reasoning-execution loop (Agent Loop) — you give it a goal, and it automatically breaks down the task, invokes tools, and executes continuously until the task is complete. You don't need to guide every step by hand.

Entry Points

Platform Entry Point
Web Visit https://www.gptbots.ai/工作区/myclaw in a browser, or click the Work menu on the left after logging in
Desktop APP After launching the APP, you enter the Work page by default; Work is also the first item in the left-side menu

Interface Layout

The Work page uses a conversation stream + overlay layout:

Screenshot location

Area Component Description
Left sidebar Side navigation Switch between Work / Search / Agent / Workflow and other menus
Top Toolbar History button, + New Task button
Center Conversation/Welcome page New sessions show the welcome page; ongoing conversations show the message stream
Bottom PromptInput Input box + node selection + model selection + working directory + attachments

Tip: On the APP, a LOCAL badge appears in the top-left corner of the window, indicating that the current node is the local node.

Welcome Page

When you create a new conversation, Work displays the welcome page:

  • Workspace Logo and brand name: customizable in Space Management → Advanced Settings
  • "Start collaborating" tagline: defaults to "Your Enterprise Work AI Partner"
  • Quick action cards: administrators can configure common task templates in the workspace, for example:
    • 📁 File organization — organize and manage files in the workspace folder
    • ✏️ Content creation — read documents in a folder and generate reports and articles
    • 📄 Document processing — process Word, PDF, and Excel documents in a folder
    • 📊 Data analysis — analyze data files in the workspace and generate reports

Click a card to start a new conversation with a preset prompt.

Input Area (PromptInput)

The input area is located at the bottom of the conversation area and provides rich input capabilities:

Feature Description
Text input Multi-line text input; Enter to send, Shift+Enter for a new line
Attachment upload Supports images, documents, audio, and other files, automatically converted into multimodal content blocks
Node selection Select the target execution node (local or remote APP node)
Model selection Switch the LLM model used for this conversation
Working directory Select which directory the Agent works in (affects the default paths for Bash, Read File, etc.)
History button Top-right corner, view all past sessions
+ New Task Top-right corner, start a new conversation

When no node is available: ● No node available is displayed below the input box. This means no APP node is online under your account, so tasks cannot be executed. Please start the desktop APP first, or ask a teammate to turn on their APP (set as an enterprise node).

Session History

Click the "History" icon in the top-right corner to open the session list panel:

Action Description
Search Filter session titles by keyword
Pin Pin important sessions to the top
Group Organize sessions using folders
Delete Remove sessions you no longer need
Rename Change the session title
Icon distinction Local conversations show a computer icon, remote node conversations show a server icon, and IM channel conversations show the corresponding platform icon

Platform Adaptation (PlatformAdapter)

The Work page achieves shared use across Web and APP through the PlatformAdapter pattern:

  • Shared layer: core components such as WorkPage, PromptInput, and SessionList reside in the claw-shared/ directory, with the source in the Web repository
  • Web adapter: webPlatformAdapter.ts — based on Redux + Gateway WebSocket
  • APP adapter: appPlatformAdapter.ts — based on Redux + coworkService (local Sidecar)

The interface users see on both ends is largely identical, but the underlying execution differs.

Quick Start: Your First Conversation

Step 1: Start a Conversation

  1. Click Work in the workspace sidebar
  2. If you are a new user, you will see the welcome page and quick action cards
  3. Enter your question or task in the input box at the bottom, and press Enter to send
  4. The Agent starts processing, and you will see the reply text and the tool invocation process in real time

Step 2: Experience Tool Invocation

Try asking the Agent to perform tasks that require tools:

Try these prompts What the Agent will do
"Show me what files are in the current directory" Invokes the List Directory tool
"Search for the latest TypeScript tutorials for me" Invokes the web-search skill
"Create a hello.py file" Invokes the Write File tool
"What's in this image?" (with an image attached) Uses vision capability to analyze the image

Step 3: Manage Sessions

  • New conversation: click the + New Task button in the top-right corner
  • Search history: "History" icon in the top-right corner → search box at the top
  • Pin important conversations: right-click a session → Pin

FAQ

The Agent replies slowly

  • Check the model currently in use (large models are slower than small ones)
  • Complex tasks involve multiple rounds of tool invocation and take more time
  • If you are using a remote node, network latency also increases response time

The Agent seems to "forget" the content from the beginning of the conversation

When a conversation gets long, the system automatically compresses the context. This is normal behavior — the system intelligently retains the most important information. If you need the Agent to remember a key piece of information, you can say "Please remember..." to store it in the memory system.

A confirmation dialog popped up during tool invocation

This is a security mechanism. P1/P2-level tools (such as Bash) require your confirmation before execution. Check the tool name and parameters, and click "Allow" once you are sure it is safe. You can adjust the permission level in Runtime Security.

Sending a message on the Web shows "No node available"

The Web cannot execute tasks on its own and requires at least one APP node to be online. Please install and start the desktop APP (APP Quick Start), or ask a team member to set their APP node as enterprise.