AI EmployeeSuper-AgentsAgent-to-AgentTutorialsPricingBlogStoryContact

Support

CellCog Cowork Guide

Connect your personal computer to CellCog and let AI agents work directly on your code, files, and projects.

What is Cowork?

Cowork (also called "Cowork on my PC") lets CellCog AI agents access your personal computer during a chat session. Instead of working only in a cloud sandbox, agents can:

  • Read and edit files on your machine
  • Run terminal commands in your local environment
  • Access your development tools, SDKs, and project dependencies
  • Work with your actual codebase in its real environment

Key Principle: You stay in control. Every action the agent wants to take on your machine requires your approval — unless you've configured auto-approve settings.

Getting Started

Step 1: Download the Desktop App

  1. Go to Cowork in the sidebar (or visit your profile)
  2. Download CellCog Desktop for your platform:
    • macOS (.dmg)
    • Windows (.exe)
    • Linux (.AppImage)

Step 2: Install and Sign In

  1. Install the downloaded app
  2. Open CellCog Desktop
  3. Sign in with your CellCog account (same credentials as the web app)
  4. The app runs in the background — you'll see a status icon in your system tray/menu bar

Step 3: Verify Connection

  1. Go to the Cowork page in the web app (sidebar)
  2. You should see "Connected" with your machine's hostname and system details
  3. If not connected, ensure the desktop app is running and you're signed in

Using Cowork in Chats

Enabling Cowork for a Chat

  1. Start a new chat or open an existing one
  2. In the message editor toolbar, click the computer icon (Cowork selector)
  3. Toggle "Cowork on my PC" to enable it
  4. Optionally set a working directory — this is the folder the agent will primarily work in (e.g., your project root)
  5. Send your message

Tip: Setting a working directory helps the agent understand your project context and affects auto-approve behavior.

What Happens Next

When the agent needs to work on your machine, it will:

  1. Propose commands (terminal commands, file reads, file edits)
  2. You'll see a command review panel appear in your chat
  3. Review each command and approve or deny it
  4. Approved commands execute on your machine via the desktop app
  5. Results are sent back to the agent, which continues working

Command Approval System

The Review Panel

When the agent sends commands for your machine, a review panel opens showing:

  • Command type: Terminal command, file read, file write, or file edit
  • Threat level: Safe, moderate, or dangerous (color-coded)
  • Summary: Brief description of what the command does
  • Full details: The exact command or file changes

Approval Actions

table
ActionEffect
ApproveExecute this specific command
DenyReject this command (optionally provide a reason)
Approve AllApprove all pending commands in the batch

Important: Commands in a batch execute in order. All commands must be decided (approved or denied) before any execute.

Threat Levels

table
LevelColorMeaningExamples
SafeGreenRoutine, no riskReading and editing files in a version-controlled working directory, listing directories
ModerateYellowSome cautionReading files outside the working directory, changes without an easy recovery path
DangerousRedNeeds attentionDeleting files, installing software, running unknown scripts, modifying system files

Auto-Approve Threshold

Cowork uses the same risk model as Browse my Chrome and connected Tools: the agent classifies every command as safe, moderate, or dangerous, and your auto-approve threshold decides which classifications run without manual review.

Accessing Settings

  1. Go to the Cowork page in the sidebar
  2. Find the Auto-approve threshold section

Threshold Levels

table
ThresholdWhat it auto-approves
NoneNothing — every command requires your approval
SafeRoutine, side-effect-light actions (e.g., reading and editing files inside a version-controlled working directory)
ModerateAlso actions that read private data or make less easily recoverable changes
DangerousEverything the agent declares — including hard-to-undo changes. Only for fully trusted workflows

Anything above your threshold shows an approval card in chat with the command, its classification, and a one-sentence reason — you approve or deny each one.

The same three risk levels and the same threshold model apply across Cowork, Browse, and Tools, so you only need to learn one risk language.

Working Directory

The working directory tells the agent where your project lives and affects auto-approve behavior.

Setting a Working Directory

  • Per chat: Set when enabling Cowork in the editor toolbar
  • Change mid-chat: Update via the Cowork selector in the editor

How It Affects Risk Classification

The working directory is a key input to how commands are risk-classified:

  • Edits inside a version-controlled working directory are typically safe (easily recoverable)
  • Reads or changes outside the working directory are classified higher (moderate or dangerous)
  • System files and shell profiles (e.g., /etc/hosts, ~/.zshrc) are always treated as dangerous to modify

Security and Privacy

What's Protected

  • Blocked paths: Sensitive locations like ~/.ssh, credential files, and system configuration are automatically blocked
  • Credential redaction: Command output is sanitized — passwords and secrets are automatically redacted
  • Encryption: All data between your machine and CellCog servers is encrypted in transit (TLS)
  • No persistent access: The agent only accesses your machine during active chat sessions with Cowork enabled

Your Controls

  • Per-command approval: Review every action before it executes
  • Auto-approve granularity: Fine-grained control over what runs automatically
  • Instant disconnect: Close the desktop app to immediately cut all access
  • Per-chat toggle: Enable/disable Cowork independently for each chat
  • Working directory scope: Limit the primary scope of agent operations

Data Handling

  • File contents accessed through Cowork may be transmitted to CellCog servers for AI processing
  • Data in transit is encrypted
  • File contents are not retained beyond the duration necessary to complete the requested operation

Desktop App Updates

  • The desktop app checks for updates automatically
  • When an update is required, you'll see an "Update Required" status on the Cowork page
  • Updates typically install automatically — if stuck, restart the app or download the latest version
  • Minimum version requirements are enforced for security and compatibility

Browse my Chrome — the Browser Companion

Cowork on my PC is the foundation; Browse my Chrome extends the same trust boundary to your real Chrome browser. With the same Desktop app installed and a small Chrome extension, your CellCog agent can use your real logged-in Chrome sessions — Gmail, GitHub, banking, internal tools — to do work that requires being authenticated as you.

Browse my Chrome uses the same approval-card model you already know from Cowork (every action shows what it does and is gated by your threshold), but instead of a working directory on disk, the trust boundary is a cyan "CellCog" tab group in your Chrome window.

See the Browse my Chrome Guide for setup and usage details.

Troubleshooting

Desktop App Not Connecting

  1. Verify the app is running: Check your system tray/menu bar for the CellCog icon
  2. Check sign-in: Ensure you're signed in with the same account as the web app
  3. Restart the app: Quit and reopen CellCog Desktop
  4. Check network: Ensure your machine has internet access
  5. Firewall: Ensure CellCog Desktop isn't blocked by your firewall

Commands Not Executing

  1. Check approval: Ensure all commands in the batch are decided (approved or denied)
  2. Check connection: Verify "Connected" status on the Cowork page
  3. Check app version: An outdated desktop app may be blocked — update if prompted

"Update Required" Message

  1. The desktop app will attempt to update automatically
  2. If stuck, quit the app completely and reopen it
  3. As a last resort, download the latest version from the Cowork page

Agent Can't Find Tools/SDKs

The agent runs commands using your system's default shell. If tools aren't found:

  1. Ensure the tool is installed and available in your shell PATH
  2. Check that the tool is initialized in your shell profile (~/.zshrc or ~/.bash_profile)
  3. Try running the command yourself in a terminal to verify

Working Directory Issues

  • Ensure the path exists on your machine
  • Use absolute paths (e.g., /Users/me/projects/app, not ~/projects/app)
  • The directory must be accessible to your user account

Frequently Asked Questions

Can the agent access my machine without my permission?

No. You must download the desktop app, sign in, and explicitly enable Cowork for each chat session. The agent cannot access your machine otherwise.

What happens if I close the desktop app mid-session?

The agent loses access immediately. Any pending commands will fail. You can reopen the app to resume.

Is my code sent to CellCog servers?

File contents may be transmitted to CellCog servers for AI processing during the session. Data is encrypted in transit and not retained after the operation completes.

Can I use Cowork with multiple machines?

Currently, one machine can be connected at a time per account.

Does Cowork work on all operating systems?

CellCog Desktop is available for macOS, Windows, and Linux.

Can the agent install software on my machine?

Only if you approve a command that installs software (e.g., npm install, pip install). Installing software is classified as dangerous, so unless your auto-approve threshold is set to Dangerous, you'll always review such commands first.

How do I completely remove Cowork access?

  1. Quit the CellCog Desktop app
  2. Optionally uninstall it from your machine
  3. No further access is possible

Can other people in my organization see my machine?

No. Cowork connects your machine to your personal CellCog account only. Organization members cannot access your machine through CellCog.

Related Guides