Overview

Nyzhi is a terminal-first AI coding agent built in Rust that reads your code, writes changes, runs tests, and verifies results — all from the command line.

TL;DR

Install with one command, set an API key, and run nyz run "Add unit tests for parse_config". Nyzhi connects to 17+ LLM providers, ships with 50+ tools, and works in both interactive chat and one-shot modes.

What Nyzhi Is

Nyzhi is an AI coding agent that lives in your terminal. You describe what you want in plain language; it figures out how to do it. It reads files, edits code, runs shell commands, executes tests, and checks that everything works — with structured evidence you can trust.

Key capabilities:

• Read — Files, diffs, semantic search, LSP diagnostics, directory trees
• Write — Create and edit files, apply patches, refactor across modules
• Run — Shell commands, tests, builds, linters
• Verify — Build logs, test output, lint results as structured evidence

It works in two modes: interactive (a rich TUI where you chat in real time) and non-interactive (nyz run "<task>" for one-shot runs from scripts or CI).

LLM Providers

Nyzhi is provider-agnostic. Connect any of 17+ LLM providers — open-source first, plus every major proprietary API:

Ollama runs models entirely on your machine — no API keys, no network calls, full privacy. See Providers for the full list and configuration.

50+ Built-in Tools

Nyzhi ships with a rich tool catalog:

• Files — read, write, edit, patch, glob, grep, move, copy, delete
• Git — status, diff, log, commit, checkout
• Shell — run commands with streaming output
• Code analysis — semantic search, LSP integration, directory trees
• Browser — fetch URLs, automate pages
• Teams — spawn coordinated sub-agents with task boards and mailbox messaging

Extend further with MCP (Model Context Protocol) servers — plug in filesystem tools, remote APIs, or custom integrations without changing Nyzhi itself.

Autonomous Execution

Autopilot runs tasks end-to-end in five phases:

1. Expansion — Turn your idea into a full specification
2. Planning — Break work into ordered steps
3. Execution — Create and modify files, run commands
4. QA — Run verification checks, fix issues
5. Validation — Confirm requirements are met

State persists between phases. If you interrupt, you can resume from where you left off.

Multi-Agent Teams

Spawn coordinated sub-agents that work together:

/team 3 Build a REST API with auth, database, and tests

A lead agent breaks the task into sub-tasks, assigns them via a shared task board, and coordinates through mailbox messaging. See Teams for details.

Sessions and Memory

• Sessions — Every conversation auto-saves as JSON. Resume, search, and export. See Sessions.
• Memory — Project-scoped and user-scoped knowledge persists across sessions. The agent learns conventions, decisions, and patterns. See Memory.

Hello World

Get from zero to your first task in under two minutes:

# 1. Install
curl -fsSL https://get.nyzhi.com | sh

# 2. Set an API key (pick one)
export OPENAI_API_KEY="sk-..."
# export ANTHROPIC_API_KEY="sk-ant-..."
# export DEEPSEEK_API_KEY="sk-..."

# 3. Run your first task
nyz run "Add a hello world function that prints 'Hello, Nyzhi!'"

Example output:

> nyz run "Add a hello world function that prints 'Hello, Nyzhi!'"

Reading src/main.rs...
Editing src/main.rs...
Running: cargo run
   Compiling my-project v0.1.0
    Finished dev [unoptimized + debuginfo]
     Running `target/debug/my-project`
Hello, Nyzhi!

Terminal UI

The interactive TUI (nyz) gives you:

• Slash commands — /help, /sessions, /resume, /autopilot, /trust, and more
• Tab completion — Commands and @-mentioned file paths
• Themes — Nyzhi Dark, Tokyo Night, Catppuccin, Dracula, Solarized, Gruvbox
• Accent colors — Copper, blue, emerald, violet, and others

See TUI for the full command reference and shortcuts.

Why Nyzhi

Next Steps

• Getting Started — Install, connect a provider, and run your first task in under two minutes
• Configuration — Global, project, and local TOML config
• Providers — Connect OpenAI, Anthropic, Gemini, DeepSeek, Ollama, and more