nyzhi

nyzhi is a terminal-first AI coding agent.

It combines a rich TUI, a one-shot CLI mode, multi-provider model support, session persistence, MCP integration, and a large built-in toolset for reading, editing, verifying, and shipping code.

Install

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

Other install paths:

# Cargo
cargo install nyzhi

# npm
npm install -g nyzhi

# from source
git clone https://github.com/nyzhi-com/code
cd code
cargo build --release -p nyzhi

Quick Start

# launch interactive TUI
nyz

# connect provider (default interactive flow)
/connect

# one-shot prompt (human-friendly text output)
nyz run "summarize this repository architecture"

# one-shot prompt (automation-friendly)
nyz exec --json "run tests and summarize failures"

Core Workflows

• nyz: interactive TUI with slash commands, history, completion, selectors, and background tasks
• /connect: default in-TUI provider setup (OAuth first, API key fallback)
• nyz run "<prompt>": non-interactive run
• nyz exec [prompt]: CI/scripting mode (reads stdin if piped)
• nyz sessions, nyz session rename, nyz export: session lifecycle management
• nyz mcp add|list|remove: MCP server configuration
• nyz teams ...: inspect and manage team metadata

Trust and Sandbox Model

Trust mode controls approval behavior:

• off
• limited
• autoedit
• full

Sandbox level controls tool execution boundaries:

• read-only
• workspace-write
• full-access

nyz exec --full_auto forces trust mode to full and sandbox to workspace-write.

Configuration Model

Primary config files:

• Global: ~/.config/nyzhi/config.toml
• Project: .nyzhi/config.toml

The runtime merges global + project config (Config::merge).
config.local.toml is supported by parsing helpers but is not currently part of the default CLI/TUI merge path.

Key sections:

• [provider] and [provider.<name>]
• [models]
• [tui]
• [agent] (trust, retry, routing, agents, verify, sharing, voice)
• [mcp]
• [shell]
• [browser]
• [memory]
• [update]
• [index]
• [external_notify]

Providers

Built-in provider IDs include:

• openai
• anthropic
• gemini
• cursor
• github-copilot
• openrouter
• groq
• together
• deepseek
• ollama
• kimi, kimi-coding
• minimax, minimax-coding
• glm, glm-coding
• claude-sdk
• codex

See docs/providers.md for auth requirements, API styles, and model notes.

Architecture Overview

                    nyzhi (CLI binary)
                           |
            +--------------+---------------+
            |                              |
        nyzhi-tui                      nyzhi-core
                                           |
                     +---------------------+---------------------+
                     |                     |                     |
                nyzhi-provider         nyzhi-auth           nyzhi-index
                     |
                nyzhi-config

Crate responsibilities:

• crates/cli: command parsing and runtime wiring
• crates/tui: terminal UX and command handling
• crates/core: agent loop, tools, sessions, teams, hooks, memory, workspace
• crates/provider: model/provider abstraction and streaming
• crates/auth: API key/OAuth/token resolution
• crates/config: schema/defaults/merge rules
• crates/index: semantic index + auto-context search

Documentation

The full local docs set lives in docs/.

• Docs index
• Architecture
• Commands
• Configuration
• Authentication
• Providers
• Tools
• TUI
• MCP
• Sessions
• Teams
• Autopilot
• Hooks
• Skills
• Memory
• Routing
• Verification
• Notifications
• Self-Update
• Building
• Releasing

Coverage and source tracing:

• Source-to-doc map

Contributing

• Contributing guide
• Support
• Security policy
• Code of conduct

License: GPL-3.0-or-later