Intro

Canopy is an AI coding agent — a fork of OpenCode with focus on token economy, model routing, and modular skills.

Let’s get started.

---

Why Canopy?

Canopy keeps everything great about OpenCode and adds:

• 🧠 Model Routing — Auto-selects the cheapest adequate model per task (DeepSeek/MiMo for scripts, Kimi for dev, Qwen for research, GLM for planning)
• 💰 Token Economy — Optimized prompts (-66%), compressed tool descriptions (-30%), aggressive compaction (-25% per turn)
• 🎓 Built-in Skills — Domain-specific packs like job-search with ATS templates, LinkedIn optimization, interview prep
• 🔧 Dual env vars — CANOPY_* and OPENCODE_* supported simultaneously for gradual migration
• 🌿 Independent — All anomalyco references removed, 32 packages renamed to @canopystack/*

---

Prerequisites

To use Canopy in your terminal, you’ll need:

1. A modern terminal emulator like:

   • WezTerm, cross-platform
   • Alacritty, cross-platform
   • Ghostty, Linux and macOS
   • Kitty, Linux and macOS

2. API keys for the LLM providers you want to use.

---

Install

The easiest way to install Canopy is through the install script.

Script

curl -fsSL https://canopy.dev/install | bash

npm

npm install -g @canopystack/canopy

Bun

bun install -g @canopystack/canopy

Homebrew

brew install mateussiqueira/tap/canopy

Source

git clone https://github.com/mateussiqueira/canopy.git
cd canopy
bun install
bun run --cwd packages/canopy --conditions=browser src/index.ts

Docker

docker run -it --rm ghcr.io/mateussiqueira/canopy

---

Configure

With Canopy you can use any LLM provider by configuring their API keys.

canopy providers

Canopy supports 75+ providers through Models.dev and local models.

---

Model Routing

Canopy automatically routes each task to the most cost-effective model:

Task Type	Recommended Model	Why
Scripts, one-liners, SQL	DeepSeek V4 Flash / MiMo V2.5	Cheapest, fast
General development	Kimi K2.6	Best quality-cost balance
Investigation, research	Qwen 3.7 Plus	Handles large context
Planning, architecture	GLM-5.1	Strong reasoning
Complex coding	Kimi K2.7 Code	Specialized for code

Configure your default model in opencode.json:

{
  "model": "kimi/k2.6",
  "small_model": "deepseek/v4-flash"
}

---

Token Economy

Canopy includes several optimizations to reduce token usage:

• System prompts: 66% shorter than stock OpenCode — same instructions, fewer tokens
• Tool descriptions: task.txt -79%, todowrite.txt -82%, edit.txt -75%
• Context compaction: Tool outputs truncated at 1,500 chars (was 2,000), less protected context
• Smart pruning: read/grep/glob outputs protected from pruning; less recent context kept verbatim

---

Skills

Skills are domain-specific instruction packs. Canopy ships with a built-in job-search skill:

{
  "skills": {
    "paths": [".opencode/skills"]
  }
}

Available skills:

Skill	Description
customize-opencode	Guide for editing Canopy’s own config
job-search	Resume/CV optimization, LinkedIn, interviews, follow-ups

The job-search skill includes:

• Resume bullet formula (Action verb + Task + Result + Tool)
• ATS optimization (keywords, section headings, format)
• LinkedIn headline and about section templates
• Interview preparation (STAR stories, common questions)
• Follow-up email templates
• Pipeline tracking (applied → screening → offer)

---

Initialize

Navigate to a project you want to work on:

cd /path/to/project
canopy

Then initialize Canopy for the project:

/init

This creates an AGENTS.md file with project context.

---

Usage

Ask questions:

How is authentication handled?

Make changes:

Add a dark mode toggle to the settings page

Plan first: Use <TAB> to switch to Plan mode, describe what you want, review the plan, then switch back to Build mode.

Undo:

/undo

---

Share

/share

Creates a shareable link to the current conversation.

---

Customize

Pick a theme, configure keybinds, set up formatters, or edit opencode.json:

{
  "$schema": "https://canopy.dev/config.json",
  "theme": "tokyonight",
  "compaction": { "auto": true, "tail_turns": 3 }
}