WorkerMill CLI

AI coding agent with multi-expert orchestration. Works with any LLM provider — Ollama, Anthropic, OpenAI, Google.

Quick Start

npx workermill

First run walks you through provider setup (Ollama auto-detected). Then just describe what you want built.

Install

# Run without installing
npx workermill

# Or install globally
npm install -g workermill
workermill

Features

• Multi-expert orchestration — Complex tasks automatically decomposed into stories, each assigned to a specialist persona (backend, frontend, devops, security, etc.)
• Any LLM provider — Ollama (local), Anthropic, OpenAI, Google. Per-persona model routing.
• 13 built-in tools — bash, read_file, write_file, edit_file, patch, glob, grep, ls, fetch, git, web_search, todo, sub_agent
• Plan mode — Read-only research phase before making changes (/plan or --plan)
• Session management — Persistent conversations with resume (--resume, /sessions)
• Cost tracking — Per-model token pricing with /cost breakdown
• Quality gates — Dangerous command warnings, permission prompts, review cycles
• Git integration — Auto-init repos, branch awareness, commit after orchestration

Usage

# Interactive mode
workermill

# Skip permission prompts
workermill --trust

# Start in read-only plan mode
workermill --plan

# Resume last conversation
workermill --resume

# Override provider/model
workermill --provider anthropic --model claude-sonnet-4-6

# Auto-revise on failed reviews
workermill --trust --auto-revise

Slash Commands

Prefix ! for direct bash: !git status, !npm test

Multi-Expert Orchestration

For complex tasks, WorkerMill automatically:

1. Classifies — Detects if the task needs multiple specialists
2. Plans — Explores the codebase, designs stories with dependencies
3. Executes — Each story assigned to a persona (backend_developer, frontend_developer, devops_engineer, etc.)
4. Reviews — Tech lead reviews all changes, with optional revision cycles
5. Commits — Stages changes and commits (with your approval)

Configuration

Config stored at ~/.workermill/cli.json (global) and .workermill/config.json (per-project).

{
  "providers": {
    "ollama": {
      "model": "qwen3-coder:30b",
      "host": "http://localhost:11434",
      "contextLength": 65536
    },
    "anthropic": {
      "model": "claude-sonnet-4-6",
      "apiKey": "{env:ANTHROPIC_API_KEY}"
    },
    "openai": {
      "model": "gpt-5.4",
      "apiKey": "{env:OPENAI_API_KEY}"
    },
    "google": {
      "model": "gemini-3.1-pro",
      "apiKey": "{env:GOOGLE_API_KEY}"
    }
  },
  "default": "ollama",
  "routing": {
    "tech_lead": "anthropic",
    "planner": "anthropic"
  },
  "review": {
    "maxRevisions": 2,
    "autoRevise": false,
    "approvalThreshold": 80
  }
}

12 Built-in Personas

backend_developer, frontend_developer, devops_engineer, qa_engineer, security_engineer, data_ml_engineer, mobile_developer, tech_writer, architect, tech_lead, planner, critic

All worker personas include production-hardened rules:

• Real services, not mocks — Docker containers for databases, caches, queues. Tests run against real services.
• Version trust — Never downgrades language/runtime versions (training data is outdated)
• Learning markers — Reports codebase discoveries with ::learning:: markers for team visibility
• Right-sized plans — Planner matches plan complexity to task complexity (1 step for simple, 3-5 for complex)
• Approval bias — Tech lead only blocks on real functional/security issues, not cosmetic preferences
• File overlap detection — Critic catches parallel merge conflicts before they happen

Custom personas can be added per-project in .workermill/personas/ or globally in ~/.workermill/personas/.

Requirements

• Node.js 20+
• An LLM provider (Ollama for local, or an API key for cloud providers)

License

MIT