I built an AI Agent that lives directly in your CLI and Desktop
Most AI coding workflows still feel disconnected.
You copy code from your editor → paste into ChatGPT/Claude → copy the answer back → repeat.
I got tired of acting like a middleman between my codebase and AI tools, so I built Mint
Mint is a unified AI desktop assistant + autonomous coding CLI designed for developers who want:
- Local-first workflows
- Agentic coding
- MCP extensibility
- Workspace-aware AI
- Transparent safety controls
What Mint can do
Unified AI CLI Agent
Mint is not just a chat wrapper.
It can:
- inspect your workspace
- search code
- read/write files
- apply patches
- execute shell commands with approval
- stream live responses directly in the terminal
Example:
mint code "find the memory leak in the auth module and patch it"
Desktop Assistant
Mint also includes an Electron desktop assistant with:
- Screen vision
- Real-time translation
- Floating assistant UI
- Live2D assistant model
- Proactive local suggestions
- Automation workflows
MCP Support
You can extend Mint using Model Context Protocol servers:
mint mcp add google-search npx --args -y @modelcontextprotocol/server-google-search
Multi-Provider Support
Mint supports:
- Gemini
- OpenAI
- Claude
- Ollama
- LM Studio
- Local OpenAI-compatible providers
Safety First
One thing I care about a lot is transparency and control.
Mint:
- asks permission before shell commands
- blocks dangerous commands
- prevents unsafe path traversal
- keeps local action logs
Why I built it
I wanted an AI assistant that actually feels integrated into the developer workflow instead of existing in a separate browser tab.
Something fast.
Something local-first.
Something extensible.
Something that feels like a real co-developer.
GitHub
GitHub: https://github.com/Pheem49/Mint
Would love feedback from other developers working on AI agents, local models, or terminal workflows