Devin AI agent

Devin AI agent is the official desktop application and CLI interface for interacting with the autonomous AI software engineer, Devin, by Cognition. The client provides a secure connection between your local development environment and Devin's isolated cloud sandbox.

Architecture and Capabilities

The client does not perform LLM inference locally. It acts as a high-performance bridge between your machine and Cognition's infrastructure via gRPC/WebSockets protocol:

• Bidirectional file system synchronization: Instant transfer of changes between the local project directory and Devin's virtual environment.
• Stream broadcasting: Real-time display of the agent's terminal, code editor, and browser with <50 ms latency.
• Local Port Forwarding: Forwarding ports from the cloud sandbox to localhost for local testing of web applications deployed by the agent.
• Session management: Initialization, pausing, resuming, and terminating agents.

---

System Requirements

• OS: Windows 10/11 (64-bit).
• Network: Stable broadband internet connection (50+ Mbps recommended for streaming the agent's browser).

---

Installation (Pre-compiled Binaries)

The most reliable installation method is using the compiled binaries from the Releases section.

Windows

1. Download the DevinClient-win-setup-v1.4.2.exe installer from the Releases section.
2. Run the .exe file.
3. Follow the installation wizard instructions. By default, the application is installed in %LOCALAPPDATA%\Programs\DevinClient.
4. Ensure the "Add devin to PATH" checkbox is ticked if you plan to use the CLI interface from PowerShell or CMD.
5. Click Install and wait for the process to complete.

---

Configuration and Authentication

You need to authenticate the client before starting work.

1. Launch the Devin application.
2. A browser window will open for OAuth2 authorization with your Cognition account.
3. After a successful login, the client will automatically retrieve and save the session token in your OS's secure storage (Keychain/Credential Manager).

Alternatively (for Headless servers or CI/CD), set the environment variable:

# Windows (PowerShell)
$env:COGNITION_API_KEY="sk-cog-your-api-key-here"

---

Quick Start (CLI)

Initialize a new project and launch the agent from the current directory:

# 1. Navigate to your project directory
cd /path/to/your/project

# 2. Initialize the Devin context
devin init

# 3. Run the agent with a specific task
devin run --prompt "Configure Webpack for a React application, add TypeScript support, and write a basic App.tsx component"

The devin run command will automatically synchronize the current folder with the agent's cloud sandbox, send the prompt, and open the TUI (Terminal User Interface) to monitor Devin's progress.

---

Support

If you find bugs in the client's behavior, please open an Issue in this repository. If the issue concerns the decision-making logic of the AI agent itself, use the "Report Agent Behavior" button in the session's graphical interface.