Now in Beta

Banana,
in your terminal.

The first vibe-coding CLI. Monkey Models bundled, native MCP, BANANA.md project instructions. One install, zero config, no API keys.

banana-cli
$ npm i -g banana-code
added 1 package in 4s
$ cd your-project && banana
Loaded BANANA.md · model: monkey/silverback
> add dark mode toggle to the settings page
Reading 12 files via LSP...
Writing src/components/DarkModeToggle.tsx
Updating src/pages/Settings.tsx
Done. Type /ripen to summarize the session.
FIRST VIBE-CODING CLI  •  NPM I -G BANANA-CODE  •  3 MONKEY MODELS BUNDLED  •  NATIVE MCP  •  BANANA.MD PROJECT INSTRUCTIONS  •  SLASH COMMANDS  •  /VOLLEY TWO-MODEL REVIEW  •  BUILD + PLAN AGENTS  •  BYOK VIA OPENROUTER  •  FIRST VIBE-CODING CLI  •  NPM I -G BANANA-CODE  •  3 MONKEY MODELS BUNDLED  •  NATIVE MCP  •  BANANA.MD PROJECT INSTRUCTIONS  •  SLASH COMMANDS  •  /VOLLEY TWO-MODEL REVIEW  •  BUILD + PLAN AGENTS  •  BYOK VIA OPENROUTER  • 
First of its kind

The first vibe-coding CLI.

Claude Code, Aider, Codex CLI? Dev tools for engineers. Banana Code is built for the way you want to ship: pick a Monkey, write what you want, hand it off when needed.

🐒

Monkey Models, no keys

Three branded tiers built in: Silverback for big jobs, Mandrill for daily work, Tamarin for surgical fixes. Bundled token. No model picker paralysis, no API keys.

🍌

Vibe-coder DNA

Same agent loop and Monkey Models as the Banana IDE and Cloud app. The CLI isn't a stripped-down dev tool. It's the same product, in the terminal.

🏓

/volley two-model review pass

Type /volley, pick a second Monkey, set the rounds. Two models go back and forth on the same problem, pushing back on each other's work and refining. Nothing else in the terminal does this.

What you get in the terminal

A full coding agent, not a chatbot.

🤖

Agentic coding loop

Sessions, tool execution, Build and Plan agents for real software work. The agent reads files, writes code, runs commands, and iterates without round-trips.

🔓

Monkey Models + BYOK

Monkey Models work with no key. Bring your own routing through OpenRouter via BANANA_OPENROUTER_KEY. Pay your provider directly, no middleman markup.

🔌

Native MCP support

MCP is built in, not bolted on. Wire any MCP server into your sessions and the agent can use those tools immediately. Plugins included.

📝

BANANA.md instructions

Drop a BANANA.md at the project root for repo-specific rules. Existing AGENTS.md, CLAUDE.md, and CONTEXT.md are read as fallbacks.

🎯

LSP-aware tools

Edits land where they should, not three lines off. The tool layer talks to your project's language server for accurate refactors and code intelligence.

👁️

Vision support

Hand the agent screenshots and images. It can read errors, understand UI, and use visual context as part of the conversation.

Banana slash commands

In-session tools without leaving the TUI.

🏓

/volley

two-model review pass

Pit two Monkeys against the same problem. They push back, refine, and challenge each other across N turns. The terminal's first.

🍌

/peel

concise help

Quick Banana help, scoped to whatever you're doing right now.

🌿

/ripen

summarize session

Compress the current session into a hand-off summary so the next run picks up where you left off.

🐒

/monkey

model status

Show Monkey Models status, the active tier, and the loaded instruction files. Plus /instructions, /banana, and /joke.

Monkey Models built in

Three branded AI tiers tuned for coding. Match the monkey to the task. No model selection paralysis.

🦍

Silverback

Goes all out. Complex refactors, architecture decisions, heavy lifting. Your technical co-founder.

🐒

Mandrill

Sharp and balanced. Feature implementation, bug fixes, code review. Gets it done right.

🐿️

Tamarin

Surgical and fast. Quick fixes, inline suggestions, boilerplate. Code only, zero fluff.

Install the CLI

The Banana Code CLI is in beta. Free during beta.

$ npm i -g banana-code

Then run banana (or bananacode) in any project. Silverback answers by default. Type /peel if you get stuck, /volley to pit two models against each other.