React
The interface
React shapes the workspace: conversations, the file explorer, and provider settings.
Your code.A touch of magic.
An AI agent that works in your project. Explore files, generate code, and approve changes before they are written.
Windows, macOS & Linux · Install from source
Work with your AI provider
The technology behind the magic.
A web interface connected to a Go backend, inside a desktop application for Windows, macOS, and Linux.
Wails
The desktop bridge
Wails connects the React interface to the Go backend and brings them together in the desktop application.
Go
The backend
Go implements the agent logic, file tools, and communication with AI providers.
Your coding workshop.
Open a folder, choose your provider, and start a conversation. Your active project is always in view.
The actual application. The top selector defines the folder you work in.
Click a screenshot to swap views.Your folder, your scope
Explicitly choose your project directory. File tools operate within that folder.
Files and conversation
Explore your project structure and talk to AI in the same workspace.
History on your machine
Sessions and messages are stored locally so you can pick up your work.
Every change. Your control.
The agent can read and create files. Before a write, review the proposed content: edit it, approve it, or reject it.
- Review the proposal
- Adjust the content
- Decide whether to write
Project / src / greeting.ts
Proposed content
export function greet(name: string) {
return `Hello, ${name}`;
}The file is only written after your approval.
A clearly defined scope.
Clear boundaries for working with AI on your code.
Active project
my-project/src/index.tspackage.jsonTools operate within this folder.
Access to the active project
Paths are validated before reads and writes. Attempts to leave the folder and access through links are blocked.
Keys in your keyring
Credentials are stored in your operating system’s native keyring, separate from conversation history.
Local history
Sessions and messages stay on your machine. AI requests are sent to the provider you select.
Local does not mean offline: generating responses requires a connection to your AI provider.
Your AI. Your choice.
Connect Google, OpenAI, Anthropic, or DeepSeek using your own API key. Switch providers from the top bar.
Gemini
GPT
Claude
DeepSeek
API usage is subject to your provider’s rates. MerlinCode does not include AI credits.
Start coding.
MerlinCode is in active development. You can try it today by cloning the repository; a published installer is not available yet.
git clone https://github.com/JoseGaldamez/merlincode.git
The GitHub guide covers Go, Node.js, and Wails requirements for your operating system.
Build something extraordinary.
Try the application, share an issue, or contribute code. MerlinCode grows with the people who use it.
What comes next
In development
- Choose models based on each task’s complexity.
- Separate the conversation list by project.
Questions, demystified.
Where can I run MerlinCode?
It works on Windows, macOS, and Linux. Currently, clone the repository and follow the requirements and steps in the README.
Do I need an API key?
Yes. Bring your own Google, OpenAI, Anthropic, or DeepSeek key. Credentials are stored in your operating system’s keyring.
Can it modify any folder?
File tools are limited to the active project you selected. Agent writes require your approval.
Does model selection depend on the task yet?
Not yet. Currently, a configured model is used for each provider. Complexity-based selection and filtering conversations by project are in development.
Does everything work offline?
History is stored on your machine, but AI responses require a connection to your provider. Request content is sent to that provider.