Unlock agentic AI capabilities on your local desktop. No enterprise subscription required.

Claude Cowork for Windows is a lightweight yet robust client that brings the powerful "Cowork" functionality (officially locked behind the expensive Claude Max tier) to local Windows environments.

By leveraging the standard Anthropic API (claude-3-5-sonnet / opus), this tool gives the neural network safe, controlled access to your local file system, allowing it to perform real work—from code refactoring to document organization—autonomously.

• About
• Key Features
• How It Works
• Installation
• Security & Sandbox
• Configuration
• Roadmap
• Disclaimer

The official Claude Cowork feature represents a paradigm shift in AI, turning a chatbot into an agent. However, it is currently restricted to high-tier subscriptions (~$100/mo). We believe agentic workflows should be accessible to all developers and power users.

Claude Cowork for Windows acts as a bridge between the Anthropic API and your OS. You provide a high-level goal in natural language, and the agent:

1. Plans a sequence of actions.
2. Executes file operations (Read, Write, Move, Delete) locally.
3. Iterates based on the results.

Cost Efficiency: You only pay for your API tokens (Pay-as-you-go). For most individual users, this is significantly cheaper than a monthly enterprise subscription.

Unlike a standard chat interface, this agent has "hands":

• CRUD Operations: Create, Read, Update, and Delete files.
• Bulk Actions: Mass renaming, format conversion, and directory restructuring.
• Deep Search: Recursive content analysis (grep-like functionality with semantic understanding).

The agent utilizes advanced reasoning loops. It does not need micro-management.

• Prompt: "Analyze all error logs in /logs from the last week and generate a summary report in Markdown."
• Action: The agent locates the files, filters by date, parses the text, and writes Report.md without further human intervention.

• No Claude Max required.
• Works with your personal API Key.
• Supports the latest models: claude-3-5-sonnet-20241022 (recommended) and claude-3-opus.

• Full support for Windows paths (backslashes, Drive letters C:/D:/).
• Native handling of text files, code, and Office documents.
• CLI Mode: Can be integrated into PowerShell scripts for automation pipelines.

The application creates a local runtime environment where the LLM can execute tools.

1. Input: You provide a prompt (e.g., "Refactor this project, split main.py into modules").
2. Reasoning: Claude generates a thought process and outputs an XML/JSON tool call.
3. Execution: The local client parses the request and performs the system call (e.g., fs.writeFileSync).
4. Feedback Loop: The client returns the operation result (Success/Error) back to the model context, allowing it to proceed to the next step.

We provide pre-compiled binaries. No Python or Node.js environment setup is required.

Navigate to the Releases page and download the latest archive for your architecture:

• claude-cowork-win-x64.exe (Standard Intel/AMD)

Extract the archive to a permanent location, e.g., C:\Tools\ClaudeCowork. (Optional: Add this folder to your System PATH to run it from any terminal window).

Run claude-cowork.exe. On the first launch, you will be prompted to enter your Anthropic API Key. The key is securely stored using the Windows Credential Manager.

Giving an AI access to your disk involves risk. We adhere to a "Safety First" philosophy.

The agent is strictly confined to the directory (and subdirectories) you specify at launch.

• Any attempt to access parent directories (../) or system folders (C:\Windows) is blocked at the binary level.

By default, the tool runs in Safe Mode:

• Read operations: Executed automatically.
• Write/Delete operations: Require user confirmation (Y/N) in the console.
• Autonomous Mode: Can be enabled via flags --autonomous for fully unattended tasks (use with caution).

• Open Source: The code is fully auditable. No hidden telemetry.
• Direct Connection: Requests go directly from your machine to api.anthropic.com. No middleman servers.

A config.yaml file is automatically generated in the application root.

core:
  model: "claude-3-5-sonnet-20241022" # Best balance of speed/intelligence
  context_window: 200000
  temperature: 0.1

safety:
  require_confirmation: true # Set to false for full autonomy
  allowed_extensions: ["*"]  # Restrict to specific types if needed (e.g. [.txt, .py])
  blocked_directories: 
    - ".git"
    - "node_modules"
    - "System Volume Information"

ui:
  theme: "dark"
  show_thought_process: true # Display the internal monologue of the AI

• Basic File System Operations (NTFS).
• CLI Argument Support.
• GUI Interface: A modern Electron-based dashboard.
• Vision Support: Ability to analyze images/screenshots within folders.
• Plugin System: Allow the agent to run local scripts (.bat, .py) safely.