Skip to content

Hackable terminal coding agent.

Work in your terminal. Choose your model. Make the agent your own.

sh
bun add -g @zaly/cli
sh
npm install -g @zaly/cli
sh
pnpm add -g @zaly/cli

Then run zaly in your project. Node.js 22.18+ · Linux & macOS

01 / Your models

Choose the model, keep the workflow. ​

Use supported providers with API keys or browser login, including ChatGPT subscription access through Codex. Switch models without leaving the conversation. Your sessions stay in your terminal, not in another browser tab.

Connect a provider →

Inside zaly

text
/login
/model
/reasoning

02 / Your workflow

Make your next prompt a command. ​

Turn a Markdown file into a reusable slash command. Add arguments, expressions, and project context. Go further with instructions, skills, and plugins—without forking the agent.

Create a command →

.zaly/commands/review.md

md
---
description: Review selected code
---

Review {{ args }} for correctness.
Explain concrete problems and suggest fixes,
but don't edit files yet.

03 / Your boundaries

Decide what needs approval. ​

Start with a permission preset, then refine the allow, ask, and deny rules. Permissions govern tool operations, not just what the prompt tells the model to do. Inspect requests and approve the work you actually want.

Understand permissions →

.zaly/config.json

json
{
  "permissions": {
    "preset": "readonly",
    "ask": ["write(*)"],
    "deny": ["bash(sudo:*)"]
  }
}

Useful today. Still alpha. APIs, configuration, and session formats can change before 1.0. These docs follow the repository; installed releases may differ. Linux and macOS are tested. Windows is untested.

Released under the MIT License.