Checking npm for latest version…

Banana Code builds fast and ships clean

A sleek AI coding command center for real projects. Watch it plan, edit, run, recover from errors, and keep momentum without dragging you through a slow workflow.

$ npm install -g @banaxi/banana-code

banana-code workspace

Fast by default

Short feedback loops, quick command approval, and model routing built for active coding sessions.

Good output

Reads the project, patches files, runs checks, and keeps improving until the change works.

Made for builders

Terminal power with a polished studio view for sessions, models, approvals, and history.

10 AI
Providers

Live speed demo

See the workflow move.

Banana Code is designed to feel immediate: ask for the change, let it work through the files, approve what matters, and get back a result you can inspect.

13.9s quick loop

10 providers

banana-code-video.mov

🧠

Multi-Provider Support

Switch instantly between Google Gemini, Anthropic Claude, OpenAI, Mistral, Deepseek, Kimi, OpenRouter, LM Studio, and Ollama Local/Cloud mid-conversation.

⚡

True Autonomy

Banana Code tries, fails, reads the errors, and fixes its own mistakes automatically with robust Self-Healing Error Loops.

🛡️

Banana Guard

AI-Powered Auto-Approve. Get the speed of YOLO mode with a smart security layer that approves safe commands and pauses risky ones.

💰

Financial Intelligence

Track exact API spend and savings in real-time. Features native 1-Hour Extended Prompt Caching for Claude to save you money.

💾

UltraMemory Context

Extreme context retention. A background AI worker continuously scans and summarizes your active chat history for deep long-term understanding.

🔌

MCP & Agent Skills

Connect to external APIs via Model Context Protocol (MCP) and teach your AI custom workflows by dropping a SKILL.md file.

How it compares

See how Banana Code stacks up against the latest features of other top AI coding tools in 2026.

Feature	Banana Code	Cursor	Claude Code
Open Source	✅ Yes	❌ No	❌ No
Multiple Providers	✅ All	Major Providers only	Claude Only
Local Models	✅ Local Ollama & LM Studio Support.	❌	❌
Security / Permissions	✅ Banana Guard, Y/N Prompts & YOLO Mode	⚠️ Y/N Prompts / YOLO mode	⚠️ Configurable Permissions
Persistent Memory	✅ UltraMemory Background Worker	⚠️ Rules & Project Context	❌ Session Context
Interactive Shell / Input	✅ Yes (Spawns persistent PTY)	✅ Yes (via built-in terminal)	Standard CLI execution
Operating Modes	✅ 5+ Specialized Modes (Plan, Ask, Security, etc.)	❌ Standard Agent	❌ Standard Agent
Model Context Protocol (MCP)	✅ Full Support + Custom Agent Skills	✅ Full Support	✅ Full Support
Auto Model Routing	✅ Yes (Fast Router Model)	❌ Manual Model Selection	❌ Claude Only
Cost	✅ Free	❌ $20 / month	❌ $20 / month

Frequently Asked Questions

Everything you need to know about Banana Code.

What exactly is Banana Code?

Banana Code is a terminal-native AI pair programmer. It's designed for developers who want the power of AI agents (like writing, debugging, and refactoring code) directly in their CLI without having to switch to a heavy IDE or web browser.

Is Banana Code free to use?

Yes! Banana Code itself is open-source and free. You only pay for the API tokens you consume from providers like OpenAI, Anthropic, Google, Deepseek, or Kimi. If you already have a ChatGPT subscription, you can connect it and use it with Banana Code. You can also use Gemini API or OpenRouter, which both provide free tiers, and local models via Ollama or LM Studio are 100% free.

Which AI models are supported?

We support almost everything: Google Gemini, Anthropic Claude, OpenAI, Mistral, Deepseek, Kimi, and any model available via OpenRouter, LM Studio, or Ollama (for local execution). You can even switch models mid-conversation.

How does "Banana Guard" work?

Banana Guard is our security layer. It uses a secondary AI check to analyze commands before they run. Safe commands (like reading a file) are auto-approved, while risky commands (like deleting directories or installing packages) will pause and ask for your permission.

Can I use it for existing projects?

Absolutely. Just run `banana` in your project root. It will index your files and be ready to help you with refactoring, bug fixing, or adding new features to your codebase immediately.

Does Banana Code respect my privacy?

Yes. Banana Code stores your API keys, chat history, and sessions locally, and the first-run install ping only saves an aggregate download count with no retained IP history. Your local CLI prompts and file contents go directly to the AI provider you configure; Banana Code does not intercept or store them. If you enable optional Banana Remote, phone-originated prompts, transient phone image attachments, AI responses, and tool request details may pass through bananacode.sh over TLS for remote chat, approvals, and session history. Read the full Privacy Policy.

Can I save money on API costs?

Yes! Banana Code has built-in Prompt Caching for Claude and shows real-time cost tracking via `/context`. You can save up to 90% on repetitive tasks by keeping the context "warm" automatically.

What are "Agent Skills"?

Skills are custom guidelines (markdown files) you can drop into your config folder. They teach the AI specific standards, like "always use TypeScript" or "follow our company's specific React patterns," which are activated automatically when relevant.

Does it support web research?

Yes. Through the `/beta` menu, you can enable DuckDuckGo integration. This allows the AI to research documentation or check real-time information on the web when it needs extra context.

Can I use it as a backend for my own apps?

Definitely. Use `banana --api` to start a headless server. You can then connect via WebSocket or HTTP to build your own custom GUI or integrations on top of the powerful Banana Code engine.

Studio Download App Privacy Policy Docs