Work Mode

Work Mode upgrades Chatbox from an AI that only talks into an AI assistant that actually does the work — it runs code, reads and writes files, calls tools, queries knowledge bases, and finishes multi-step tasks on its own. This page starts from the basics and goes deep, covering every capability and setting along the way.

Two Modes: Chat vs Work

There is a robot icon button on the input box toolbar — that is the mode switch. The small label next to it tells you which mode you are in.

• Chat Mode — a clean text conversation. Best for Q&A, writing, translation, explanations, and brainstorming. The robot icon is gray.
• Work Mode — on top of chatting, the AI gains a full toolset: running code, reading/writing files, calling MCP, loading skills, and querying knowledge bases. The robot icon turns brand blue.

How It Works

In Work Mode the AI no longer just thinks once and answers. It runs a loop: think about what to do, call a tool, read the tool result, decide the next step — and repeat until the task is done, then give the final answer.

Every thought and tool call shows up as a live timeline step inside the reply. Steps are collapsed by default, showing just one line — tool name · summary · duration. Click any step to expand it and inspect the call's arguments and result. When the task finishes, a "Worked for …" header appears above the timeline — click it to fold the whole process away and read only the final answer.

{ "path": "sales_2026.csv" }

Switching Modes & Smart Switching

Hover over the robot icon to open the control panel. The Mode section at the top has just two buttons — Chat Mode and Work Mode. Pick one, and that is the mode you use.

Not sure which to pick? Right below the mode buttons there is a Smart Switching toggle to help you decide. It is on by default and appears only while Chat Mode is selected and you have not sent your first message yet:

• When it is on, the AI quickly judges whether your first message is better handled in Work Mode.
• If so, it does not switch on its own — it shows a "Work Mode suggested" card in the message flow, with a one-line reason written by the AI, and you decide with one click.
• If not, it silently handles the message as a normal chat and will not ask again in this conversation.

The Control Panel

The control panel gathers everything in one place. From top to bottom it has three sections — Mode, Built-in, and Extensions. Hover over a row and a sub-panel slides out on the right for further settings (pick a search provider, change the approval level, select a knowledge base, and so on). The capabilities are only clickable in Work Mode.

Built-in Capabilities

Web Search

Lets the AI search the web and read page content. You can choose a provider: Chatbox AI (built-in, requires a license), Bing Search, or Tavily (needs your own API key). Web search works in Chat Mode too — just toggle the globe icon on the input box toolbar; once you are in Work Mode, it is managed from the control panel instead.

Code Execution

The heart of Work Mode. The AI can run Node.js or Bash in a local sandbox to process data, generate files, and convert formats, with helpers for reading, writing, editing, and searching files, plus exporting downloadable files.

Code Execution has an important safety option — the approval level:

• Approve (default) — asks before running commands or changing files. Safe and controllable.
• Full Access — skips approval prompts for commands and file changes; the AI runs automatically. Fast but risky; use it only when you fully trust the task.

Extensions

• Skills — expert playbooks that teach the AI how to do a class of tasks more professionally (based on the agentskills.io spec, compatible with Claude Code skills). The AI loads a skill's full instructions only when needed.
• MCP — connect the AI to external tools and services. Chatbox ships several cloud MCP servers (Fetch, Sequential Thinking, EdgeOne Pages, arXiv, Context7 — Premium), and you can add your own.
• Knowledge Base — bind a knowledge base and the AI can semantically search it and read documents to answer from your own materials (RAG).
• Working Directory (Desktop; not yet on Windows) — grant the AI access to a real local folder, and it can read and write inside it without per-action approval.

Safety & Control

Work Mode can act, yet it always keeps you in control:

npm run build

Real-World Workflows

Data Analysis

Upload a CSV/Excel and ask the AI to compute metrics, find patterns, and produce results. It reads the file, runs code to crunch numbers, writes the conclusion into the answer, and exports the summary table to the Artifacts section at the bottom of the message, ready to preview or save.

// Parse the CSV and sum revenue per month
const rows = parseCsv('sales_2026.csv')
const byMonth = groupSum(rows, 'month', 'revenue')
console.table(byMonth)

Month      Revenue
2026-09   $286,400
2026-10   $335,700
2026-11   $412,900
2026-12   $368,200

Coding

Add your project folder to the Working Directory and the AI can browse code, locate issues, edit files, and run scripts — like a pair programmer sitting next to you.

- <button className="login-btn">
+ <button className="login-btn" onClick={handleSubmit}>

PASS  src/pages/Login/LoginButton.test.tsx
Tests: 12 passed, 12 total

Settings Cheat Sheet

FAQ

Why don't I see the robot icon? Work Mode is desktop only. If the icon is gray and disabled, the current model usually does not support tool use — switch to a tool-capable model (such as Claude or GPT).

Can I switch modes after the chat starts? From Chat Mode you can switch to Work Mode at any time. The reverse is not possible — once you send a message in Work Mode (or accept the suggestion card, or load a skill), the mode locks, and you need a new chat to return to Chat Mode.

Is my data safe? Under the default approval level, running commands and modifying real files always ask for your consent, and code runs in a local sandbox. Only when you turn on Full Access or add a folder to the Working Directory do those actions skip approval.