COCO — Autonomous Coding Agent

COCO is an AI coding agent with a quality convergence loop that iterates until your code meets a configurable quality threshold. It runs entirely in VS Code's integrated terminal — no custom webview, no context switching.

Features

Multi-provider LLM support — Anthropic Claude, OpenAI GPT, Google Gemini, Kimi, Ollama, LM Studio, Groq, Mistral, DeepSeek, OpenRouter, Together AI, HuggingFace
Quality convergence — the agent self-reviews its output and iterates until it reaches a score ≥ 85/100
Terminal-first UX — COCO lives in VS Code's integrated terminal, just like the CLI
REPL with slash commands — /model, /provider, /intent, /mcp, and more
MCP support — connect Model Context Protocol servers for extended tooling
Persistent sessions — reuses the same terminal panel across window reloads
Zero config to start — just install and run; configure your provider API key when prompted

Requirements

The COCO CLI must be installed on your system:

npm install -g @corbat-tech/coco

Verify the installation:

coco --version

Usage

Open COCO

Method	Action
Status bar	Click `$(robot) COCO` in the bottom-left
Keyboard	`Ctrl+Shift+O` / `Cmd+Shift+O`
Command Palette	`COCO: Open COCO Agent`
Editor title bar	Click the robot icon

Start a new session

Open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P) and run:

COCO: New COCO Session

This destroys the current terminal and starts a fresh COCO session.

Inside the COCO REPL

Once the terminal opens, you interact with COCO using natural language or slash commands:

> build me a REST API with authentication

/provider          — change LLM provider
/model             — change model
/intent            — set a persistent goal
/mcp               — manage MCP servers
/help              — list all commands

Configuration

Open VS Code Settings (Ctrl+, / Cmd+,) and search for COCO.

Setting	Default	Description
`coco.cliPath`	`coco`	Path to the `coco` binary. Set an absolute path if it's not on `$PATH`.

Example — if you installed coco locally:

{
  "coco.cliPath": "/usr/local/bin/coco"
}

Supported Providers

Configure your preferred provider on first launch or via /provider:

Provider	Models
Anthropic	Claude Sonnet 4.6, Opus 4.6, Haiku 4.5
OpenAI	GPT-5.3 Codex, GPT-4.1, o4-mini
Google Gemini	Gemini 2.5 Pro, 2.0 Flash
Kimi / Kimi Code	kimi-for-coding
Ollama	Any locally downloaded model
LM Studio	Any locally downloaded model
Groq	Llama, Mixtral (ultra-fast)
Mistral	Codestral, Mistral Large
DeepSeek	DeepSeek Coder V3
OpenRouter	100+ models via one API key
Together AI	Open-source models
HuggingFace	Open models with free inference

How Quality Convergence Works

COCO's COCO loop (Converge → Orchestrate → Complete → Output) evaluates output across 12 quality dimensions on every iteration:

Correctness · Security · Performance · Maintainability · Test coverage
Documentation · Type safety · Error handling · Code style · Complexity
Dependencies · Modularity

The loop runs until the quality score reaches ≥ 85/100 or the maximum iteration count is hit. You get production-ready code, not just a first draft.

Troubleshooting

coco: command not found

The CLI is not on your $PATH. Either:

Install globally: npm install -g @corbat-tech/coco
Or set coco.cliPath to the absolute path of the binary

Terminal opens but COCO exits immediately

Check that Node.js ≥ 22 is installed: node --version

Want to use a local model?

Install Ollama (https://ollama.com), pull a model, then run /provider inside COCO and select Ollama.

COCO Code

corbat-tech