Description
Nyra is an open-source AI agent that operates directly on your personal computer, emphasizing user control and data privacy. It's designed to understand projects, modify code, execute commands, and conduct web research, but critically, it always seeks your explicit approval before performing any write operations or command executions. This "asks first, then acts" approach ensures that you, the user, remain in charge of every action the agent takes.
Read More
The agent supports various operating systems, including Windows, macOS (Apple silicon and Intel), and Linux (x64 and arm64), with specific installation instructions provided for each. It offers a flexible interface, accessible via terminal, a desktop preview application, and a web-based phone interface, allowing you to manage and approve tasks from multiple devices. The desktop and phone interfaces share the same session and approval queues, ensuring a consistent experience.
Key features of Nyra include its commitment to local-first data storage, with all session data and settings residing in a `~/.nyra` directory on your machine. API keys are stored securely in an encrypted vault or a permissions-restricted file, and never transmitted in plain text. The system also incorporates a robust undo mechanism, saving checkpoints before every write operation, allowing users to revert changes if needed. Commands are executed within a sandbox environment (bwrap on Linux, with macOS and Windows sandboxing planned for future releases) to enhance security.
Nyra is free and open-source under the MIT License, meaning the software itself incurs no cost. Users are responsible for any fees associated with model usage from third-party providers, though Nyra also supports local models like Ollama or reusing existing Codex sign-ins. It supports a wide array of models, including DeepSeek (default), xAI Grok, OpenAI, Qwen, Kimi, Zhipu GLM, Gemini, OpenRouter, Mistral, Groq, Together, and any OpenAI-compatible custom endpoint. Future plans include agent collaboration features, allowing users to work with friends' agents while maintaining granular control over permissions and data access.
- CategoryAI Agent Builders
- Pricing ModelFree
- DeploymentLocal
- PlatformsWindows, macOS, Linux
Nyra's Core Features
Open-source AI agent
Runs locally on your computer
Requires user approval before writes or commands
Reads projects and understands context
Edits code and runs commands
Performs web searches for verification
Supports Windows, macOS, and Linux
Offers terminal, desktop, and phone interfaces
Local-first data storage in ~/.nyra
Secure storage of API keys (encrypted vault or restricted file)
Never transmits plain text keys
Undo mechanism with checkpoints before writes
Sandboxed command execution (bwrap on Linux)
Supports multiple AI model providers
Fluent in Chinese and English (next release)
How to use Nyra?
Install Nyra: Download the command-line application for your operating system (Windows, macOS, or Linux) or use the PowerShell installer for Windows.
Run Setup: Open a new terminal or PowerShell window and execute `nyra setup` to complete the initial configuration.
Access Interface: Use `nyra ui --open` for the desktop interface in your browser, `nyra` for the terminal, or `nyra ui --pair` to connect your phone.
Approve Actions: Review proposed file changes (diffs) or commands and explicitly approve them before Nyra executes them.
Manage Data: All data and settings are stored locally in `~/.nyra`; manage your API keys securely within the application or via environment variables.
Update Nyra: Re-run the installation command to update to the latest version.
Nyra's Use Cases
- Code Editing & Refactoring
- Automated Command Execution
- Web Research & Verification
- Project Understanding
- Secure Local Automation
- Remote Task Approval
- Version Control Assistance








