I Built an Open-Source CLI AI Coding Assistant in Go (and it costs $3/month)

Claude Code Too Expensive? Meet Gokin — An Open Source Alternative in Go

I love Claude Code. It’s powerful, fast, and genuinely helpful. But at ~$100/month, those usage limits run out faster than I’d like.

I looked into Chinese AI alternatives like GLM-4, but I wasn’t comfortable giving their official CLIs full access to my entire codebase.

So I built my own.

What is Gokin?

Gokin is a CLI AI coding assistant written in Go. It connects directly to:

• Google Gemini (Free tier available via AI Studio).
• GLM-4 (Cost-effective, typically ~$3/month for heavy usage via API).

My Hybrid Workflow

Instead of burning expensive Claude tokens on boilerplate, I split my tasks:

• Gokin (GLM-4 / Gemini): Drafting code from scratch, bulk file operations, and repetitive tasks (80% of the work).
• Claude Code (Opus): Polishing, complex architectural changes, and deep code reviews (the critical 20%).

Features That Actually Matter

40+ AI Tools

Gokin isn't just a text generator; it interacts with your environment:

# File operations
> Create a REST API handler for user authentication

# Search
> Find all functions that handle database connections

# Git integration
> /commit -m "feat: add user validation"
> /pr --title "Add validation feature"

Semantic Code Search

Instead of wrestling with grep or regex, search by intent:

• "Where is error logging implemented?"
• "Find code related to JWT validation."
• "Show all payment processing functions."

Gokin uses embeddings to understand your codebase conceptually, making navigation much faster.

Multi-Agent System

Different agents handle different specialized tasks:

Security First

Privacy was my primary motivator. Gokin ensures:

• Secret Redaction: API keys, tokens, and passwords are automatically stripped before reaching the AI.
• Command Sandboxing: Potentially dangerous operations are blocked or require manual approval.
• Local Processing: Your raw code stays on your machine; only relevant context is sent to the API.

Cost Comparison

🚀 Installation

Choose the method that best fits your workflow:

1. Direct Go Install (Recommended)

If you have Go 1.23+ installed, this is the fastest way. The binary will be installed to your $GOPATH/bin.

go install github.com/ginkida/gokin/cmd/gokin@latest

2. Build from Source

Use this method if you want to inspect the code or contribute to development.

# Clone the repository
git clone https://github.com/ginkida/gokin.git
cd gokin

# Build the binary
go build -o gokin ./cmd/gokin

# (Optional) Move to your system PATH for global access
sudo mv gokin /usr/local/bin/

3. Shell Configuration

To ensure you can run gokin from any directory, make sure your Go bin path is in your environment variables. Add this to your ~/.zshrc or ~/.bashrc:

export PATH=$PATH:$(go env GOPATH)/bin

🛠 System Requirements

🔍 Quick Verification

After installing, check if everything is set up correctly by running:

gokin /doctor

This built-in command will verify your environment, API keys, and system permissions.

What’s Next?

• Support for more AI providers (OpenAI, local models via Ollama).
• Advanced planning algorithms (Refining existing MCTS and A* implementations).
• A dedicated plugin system.

Check it out on GitHub: github.com/ginkida/gokin

Star the repo if you find it useful! PRs are always welcome.

Have you built your own AI tools? What does your AI-assisted workflow look like? Let me know in the comments!