Local-first, multi-agent AI coding in your terminal.

Getting Started · Usage · Configuration · Benchmarks

What is cmdr?

cmdr is an AI coding assistant that runs entirely on your machine using Ollama. No API keys, no cloud, no data leaves your laptop.

npm install -g cmdr-agent
cmdr

Contents

• Screenshots
• Highlights
• Quick Start
• Documentation
• Contributing
• License

Screenshots

Highlights

| | Feature | | |---|---|---| | 🔒 | Local-first | All inference on your hardware via Ollama | | 🤖 | Multi-agent teams | Code review, full-stack, security audit presets | | 🛠 | 18 built-in tools | Files, grep, glob, bash, git, web fetch, think, browser, RAG search | | ✅ | Human-in-the-loop | Approve, deny, or always-allow each tool call | | 🧠 | Context compaction | Multi-stage strategy keeps long conversations in bounds | | 🔌 | Plugins & MCP | Extend with npm modules or Model Context Protocol servers | | 💾 | Session persistence | Auto-save, resume, --continue, checkpoints, branches | | ↩️ | Undo | /undo reverts any file change the agent made | | 📊 | Token tracking | /cost for per-session usage breakdown | | 📁 | Project awareness | Reads CMDR.md for project-specific instructions | | 🌐 | HTTP API | cmdr serve exposes REST + SSE endpoints | | 🖼️ | Vision | Attach images to prompts with --image | | 🎯 | Effort levels | --effort low\|medium\|high\|max controls reasoning depth | | 🔍 | RAG indexing | Index codebases and search with embeddings | | 🌿 | Branching | Fork, switch, and merge conversation branches | | ✅ | Checkpoints | Save and restore conversation snapshots | | 🐾 | Buddy | Deterministic ASCII companion that levels up as you code | | 👁️ | Daemon mode | Watch files and run commands on change | | 🌐 | Browser agent | Optional Playwright-based browser automation | | 🧩 | VS Code extension | @cmdr chat, inline completions, code actions |

Quick Start

# 1. Install Ollama — https://ollama.ai
ollama pull qwen3-coder:latest

# 2. Install cmdr
npm install -g cmdr-agent

# 3. Go
cmdr

# One-shot mode
cmdr "fix the failing tests"

# Pick a model
cmdr -m llama3.1:8b

# Multi-agent review
cmdr --team review

# Attach an image
cmdr --image screenshot.png "what's wrong with this UI?"

# Control reasoning depth
cmdr --effort max "architect a plugin system"

# Quick mode (alias for --effort low)
cmdr --fast "rename this variable"

# Start HTTP API server
cmdr serve --port 3120 --model qwen3-coder:latest

# Daemon mode — watch and react
cmdr daemon start --watch src --on-change "npm test"

Documentation

| Page | Description | |------|-------------| | Getting Started | Installation, first run, verify | | Usage | CLI flags, slash commands, built-in tools | | Multi-Agent Teams | Team presets and orchestration | | Plugins & MCP | Plugin system and MCP integration | | Configuration | Config files, env vars, permissions, CMDR.md | | Benchmarks | Eval suite and HumanEval results | | Architecture | Project structure and development setup |

What's New in v3

• cmdr serve — HTTP API with REST and SSE streaming endpoints
• Effort levels — Replace --think with --effort low|medium|high|max
• Vision input — --image / /image for multimodal prompts
• Conversation branching — /fork, /switch, /merge branches
• Checkpoints — Save and restore conversation snapshots
• RAG indexing — /index a codebase, /search with embeddings
• Code review — /review with staged, range, and path options
• Daemon mode — cmdr daemon start --watch src --on-change "npm test"
• Browser agent — Playwright automation with --browser
• Buddy system — ASCII companion pet that levels up (disable with --no-buddy)
• VS Code extension — @cmdr chat participant, inline completions, code actions
• SDK exports — Use cmdr as a library: import { Agent, createAdapter } from 'cmdr-agent'

Contributing

See CONTRIBUTING.md.

License

MIT — Reyyan Ahmed