Clew CLI Docs

Open-source, local-first AI coding agent CLI. Runs on your own hardware using your own API keys — supporting 29 LLM providers.

---

Introduction

Clew Code is an autonomous coding agent that runs directly in your terminal. It features a complete Ink-based React REPL interface, automatic context compaction, SQLite-backed long-term memory, LAN peer discovery for multi-machine swarms, and modular tool integrations.

---

Installation

Choose your preferred installation method below:

One-liner (Recommended)

This script installs Bun automatically if missing, then installs the Clew Code CLI.

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/ClewCode/ClewCode/main/scripts/install.sh | bash

Windows (PowerShell as Admin):

irm https://raw.githubusercontent.com/ClewCode/ClewCode/main/scripts/install.ps1 | iex

With npm

Requires Node.js 18+ and Bun 1.3+.

npm install -g clew-code

From Source (Bun)

# Clone the repository
git clone https://github.com/ClewCode/ClewCode.git
cd ClewCode

# Install dependencies & build
bun install
bun run build

# Link globally
bun link

---

First Run

Simply type clew in your terminal inside any project directory to launch the interactive REPL:

clew

To resume a past conversation or launch a specific session, use the session flags:

# Resume your last active session
clew --resume last

# Start a plan-mode session (bypasses prompt confirmations)
clew --plan

# Pass a one-shot prompt without entering REPL
clew -p "summarize CHANGELOG.md"

---

29 Providers

Clew Code is provider-agnostic, supporting 29 native LLM engines. Setup your API key as an environment variable before launching the CLI:

Provider	Environment Variable
OpenAI	OPENAI_API_KEY
Google Gemini / Code Assist	GOOGLE_API_KEY
DeepSeek	DEEPSEEK_API_KEY
Groq	GROQ_API_KEY
OpenGateway	OPENGATEWAY_API_KEY
Ollama (Local)	OLLAMA_HOST (defaults to http://localhost:11434)
Clew Gateway	CLEW_GATEWAY_KEY

Switch between providers or models at any time inside a session using the /model slash command.

---

Persistent Memory

Clew Code contains a MiMo-inspired, SQLite-backed long-term memory system. It automatically:

• Aggregates factoids, style preferences, and architectural decisions about your codebase.
• Ranks memory importance and frequency of access.
• Injects relevant context chunks back into the prompt cycles.
• Stores human-readable markdown summaries locally under ~/.clew/memory/.

---

LAN Peer Swarm

With zero configuration, Clew Code instances discover each other on the local network (LAN) using UDP multicast. You can:

• Sync project databases and memories across developer machines.
• Delegate execution tasks to other machines on the LAN.
• Broadcast bash commands to run concurrently on all peers (Swarm execution).

---

Execution Layers

Tasks are distributed across several layers depending on your workflow:

• Agent: The root agent session coordinating permissions, UI, and active memory context. Custom agents live in .clew/agents/*.md.
• Subagent: Focus-oriented, short-lived child processes spawned in the background (typically read-only) to solve specific sub-tasks without cluttering your main session.
• Peer: Network-connected nodes coordinating LAN swarms.
• ExecAgent: External CLI/Codex process delegation for running separate specialized model loops (formerly "Process Peer").
• Background Agent: Persistent task queue with lease-based concurrency, cron scheduling, and daemon mode.

---

Slash Commands

Control the session directly using terminal slash commands:

Command	Purpose
/model	Switch provider or model
/status	Provider, session, context info
/doctor	Diagnostics and provider health checks
/context	Active context usage
/compact	Compress conversation history + extract memories
/goal	Track and verify overall task completion
/mcp	MCP server management
/code-review	Review changed files for bugs
/plugin	Plugin and hook management
/agent	Background agent dispatch
/agents	TUI Agent dashboard
/peer	LAN peers: share, discover, swarm, dashboard
/remote	WebSocket remote control
/daemon	Autonomous daemon dashboard
/task	Scheduled tasks
/memory	SQLite memory system management
/tasks	Curated task checklist
/effort	Set reasoning effort
/rule	View / toggle project rules
/workspace	Link / unlink / list project repos
/cd	Change working directory
/diff	Show git diff
/files	Explore project files
/branch	Git branch operations
/fork	Fork a conversation
/stats	Session statistics
/guardian	Auto-review mode using secondary LLM
/approve	Override guardian denials
/pr	GitHub PR lifecycle
/voice	Voice input
/buddy	Companion card
/team	Team dashboard
/bg	Background sessions
/plan	Plan mode
/vim	Vim keybindings
/research	Research dossier management
/deep-research	Deep research with multi-source synthesis
/workflow	Multi-step workflow automation
/rewind	Undo last response
/upgrade	Check for updates
/session	Session management
/theme	Theme switcher
/skills	List and manage skills
/ultracode	UltraCode reasoning mode
/cost	Token cost summary
/thinkback	Browse past agent thoughts
/passes	View test/code passes
/hooks	Manage lifecycle hooks
/export	Export conversation
/sandbox-toggle	Toggle sandbox mode
/insights	Generate session analysis report

Profile is set via settings.json ("profile": "personal") — no slash command.

---

Configuration

Settings files are stored in .clew/ at the project root:

• .clew/settings.json — Shared project settings (checked into version control)
• .clew/settings.local.json — Private/local overrides (gitignored, never commit)

Also supports ~/.clew/ for global defaults.

Example settings.json

{
  "defaultProvider": "deepseek",
  "defaultModel": "deepseek-reasoner",
  "theme": "dark",
  "autoApprove": {
    "readFiles": true,
    "writeFiles": false,
    "runCommands": false
  },
  "hooks": {
    "PostToolUse": "echo 'tool used: $TOOL_NAME'"
  },
  "permissions": {
    "allow": ["Bash", "Read", "Edit", "Write", "Glob", "Grep"],
    "deny": []
  }
}