Kode

Your AI-Powered Terminal Coding Companion

中文文档 · Contributing · Documentation · Releases

---

Understand your codebase · Edit files · Execute commands · Orchestrate workflows

Table of Contents

• Highlights
• Installation
• Quick Start
• Multi-Model Collaboration
• Agents & Subagents
• Skills & Plugins
• MCP Extensions
• Permissions & Security
• Configuration
• Development
• License

Highlights

Intelligent Coding Multi-Model Orchestration — Combine 20+ AI models, each excelling at different tasks Expert Consultation — @ask-model-name for specialized analysis Agent Delegation — @run-agent-name for task orchestration Smart Completions — Fuzzy matching with 7+ algorithms

Developer Experience Zero-Config Start — Works out of the box with any OpenAI-compatible endpoint AGENTS.md Standard — Compatible with 60k+ open-source projects Rich Terminal UI — Syntax highlighting, image support, inline editing Extensible — Skills, plugins, MCP servers, custom agents

Note

Security: Kode runs in YOLO mode by default for maximum productivity. Use kode --safe to enable permission checks when working with critical files.

Model Advice: Use agentic models designed for autonomous task completion (not Q&A-focused models like GPT-4o) for best results.

Installation

npm install -g @shareai-lab/kode

🇨🇳 China Mirror / Additional Options

# China mirror
npm install -g @shareai-lab/kode --registry=https://registry.npmmirror.com

# Dev channel (latest features)
npm install -g @shareai-lab/kode@dev

Kode bundles per-platform ripgrep and native binaries via optionalDependencies. If installed with --no-optional, install system rg or set KODE_RIPGREP_PATH.

Standalone Binary (no npm)

Download Bun-compiled binaries from GitHub Releases. See docs/binary-distribution.md.

After installation, use any of these commands:

Command	Description
kode	Primary command
kwa	Kode With Agent
kd	Ultra-short alias

Quick Start

Interactive Mode

kode

One-Shot Mode

kode -p "explain this function" path/to/file.js

ACP Mode (Agent Client Protocol)

kode-acp          # stdio JSON-RPC for Toad/Zed clients

Keyboard Shortcuts

Shortcut	Action
Enter	Submit message
Option+Enter	Insert newline
Option+M	Cycle active model
Option+G	Open message in $EDITOR
Ctrl+V	Attach clipboard image (macOS)

Slash Commands

/model          Change AI model settings
/config         Open configuration panel
/agents         Manage subagents
/plugin         Manage skills & plugins
/output-style   Switch output behavior
/cost           Show token usage & costs
/clear          Clear conversation
/help           Show all commands

Multi-Model Collaboration

Unlike single-model tools, Kode enables true multi-model orchestration — assign the right model to the right task.

Architecture

┌─────────────────────────────────────────────────┐
│                  ModelManager                     │
├──────────┬──────────┬───────────┬───────────────┤
│   main   │   task   │  compact  │     quick     │
│ (primary)│(subagent)│(summarize)│  (utilities)  │
└──────────┴──────────┴───────────┴───────────────┘
         │                │                │
    Main Agent      SubAgents       Expert Consult

Model Pointers — Configure defaults for each role via /model:

Pointer	Purpose
main	Primary conversation model
task	SubAgent / delegation model
compact	Context compression near window limit
quick	Fast operations & utilities

Shareable Config (YAML)

kode models export --output kode-models.yaml   # Export (no plaintext keys)
kode models import kode-models.yaml            # Import (merge)
kode models import --replace kode-models.yaml  # Import (replace)
kode models list                               # List profiles + pointers

Workflow Examples

# Architecture — use reasoning models
"Use o3 to design the message queue architecture"

# Implementation — use coding models
"Use Qwen Coder as subagent to refactor these three modules in parallel"

# Expert consultation — consult a specialist
"Ask Claude Opus 4.1 about this memory leak"

# Model switching — Option+M or specify inline
"Switch to Kimi k2 for code review"

Key Capabilities

Feature	Kode	Single-Model CLI
Models Supported	Unlimited	One
Live Switching	Option+M	Restart required
Parallel SubAgents	Yes	No
Per-Model Cost Tracking	Yes	No
Expert Consultation	@ask-*	Not available

Agents & Subagents

Agents are reusable templates for task delegation and orchestration.

# Manage
/agents                          # Interactive UI
kode agents validate             # Validate templates

# Run
@run-agent-reviewer ...          # Via @ mention
Task(subagent_type: "reviewer")  # Via tool call

Agent file (.kode/agents/reviewer.md):

---
name: reviewer
description: "Review diffs for correctness, security, and simplicity"
tools: ["Read", "Grep"]
model: inherit
---

Be strict. Point out bugs and risky changes. Prefer small, targeted fixes.

Sources: .kode/agents (project) → ~/.kode/agents (user) → plugins → --agents flag.

Model field accepts: inherit, pointer names (main|task|compact|quick), profile names, or provider:modelName.

Skills & Plugins

Install from Marketplace

kode plugin marketplace add owner/repo
kode plugin install document-skills@anthropic-agent-skills --scope user

Use Skills

Run as slash commands (/pdf, /xlsx) or let Kode invoke them automatically.

Create a Skill

Create .kode/skills/<name>/SKILL.md:

---
name: my-skill
description: Describe what this skill does and when to use it.
allowed-tools: Read Bash(git:*) Bash(jq:*)
---

# Skill instructions here

See docs/skills.md for full reference.

MCP Extensions

Connect to Model Context Protocol servers to extend Kode's capabilities.

kode mcp add              # Add server
kode mcp list             # List connected servers
kode mcp remove <name>    # Remove server

Config file (.mcp.json in project root):

{
  "my-server": { "type": "sse", "url": "http://127.0.0.1:3333/sse" }
}

Permissions & Security

Mode	Behavior
Default (YOLO)	Skips permission prompts for speed
kode --safe	Requires approval for writes & commands
Plan Mode	Read-only until plan is approved

System Sandbox (Linux)

With --safe or KODE_SYSTEM_SANDBOX=1, Bash commands run inside a bwrap sandbox (network disabled by default).

Network & Privacy

• No telemetry by default
• Network requests only for: model API calls, web tools, plugin downloads, optional update checks

Configuration

Scope	Location
Global config	~/.kode.json (or $KODE_CONFIG_DIR/config.json)
Project settings	.kode/settings.json, .kode/settings.local.json
MCP servers	.mcp.json or .mcprc
Agents	.kode/agents/*.md
Skills	.kode/skills/*/SKILL.md

AGENTS.md Support

Kode discovers instruction files from repo root to CWD:

• Prefers AGENTS.override.md > AGENTS.md (one per directory)
• Concatenated root → leaf (32 KiB cap, override with KODE_PROJECT_DOC_MAX_BYTES)
• Legacy CLAUDE.md / .claude/ compatibility included

Docker

git clone https://github.com/shareAI-lab/Kode.git && cd Kode
docker build --no-cache -t kode .

# Run in your project
cd your-project
docker run -it --rm \
  -v $(pwd):/workspace \
  -v ~/.kode:/root/.kode \
  -v ~/.kode.json:/root/.kode.json \
  -w /workspace \
  kode

Development

Requires Bun for development.

# Setup
git clone https://github.com/shareAI-lab/kode.git && cd kode
bun install

# Dev / Build / Test
bun run dev
bun run build
bun test

License

Apache 2.0 — Use freely in personal, commercial, and enterprise projects.

Acknowledgements

• Some code from @dnakov's anonkode
• UI inspiration from gemini-cli
• System design learned from upstream agent CLIs

---

Documentation · Report Issues · Discussions

_{From July through December 2026, Kode is undergoing a major maintenance and refactoring phase.
Follow releases and announcements for the refreshed roadmap.}