agentic-workforce
v0.1.1
Published
A multi-agent AI workforce that plans, delegates, and executes tasks through a company hierarchy of LLM agents
Downloads
287
Maintainers
Readme
Agentic Workforce
A multi-agent AI CLI workforce that plans, delegates, and executes tasks through a simulated company hierarchy of LLM agents.
$ agentic-workforce "Refactor the main service to use dependency injection"Each agent operates in a PLAN → ACT → OBSERVE → REFLECT → VERIFY loop. Tasks flow from CEO → VP Engineering → Engineering Managers → Software Engineers → QA, with cost tracking and escalation at every level.
Quick Start
# Install globally
npm install -g agentic-workforce
# First run starts the configuration wizard
agentic-workforce
# Or jump straight in with a goal
agentic-workforce "Review the codebase and suggest improvements"
# Shorter alias
awf "Deploy the staging environment"Configuration
On first run you'll be guided through the provider setup wizard. You can configure:
- Multiple providers — OpenAI, Anthropic, Groq, Google, Ollama, Cerebras, OpenRouter, or custom
- Tier mapping — assign different models to different agent tiers (frontier/mid/cheap-free)
- Cost cap — daily spending limit
- Custom models — enter any model ID for any provider
# Re-run the configuration wizard
agentic-workforce --configureTier Model
| Tier | Roles | Recommended | |------|-------|-------------| | Frontier | CEO, CTO | GPT-4o, Claude Sonnet 4 | | Mid | VPs, Directors, Managers | GPT-4o-mini, Claude Haiku | | Cheap-Free | Engineers, QA, Support | Qwen 2.5 Coder (Ollama), Gemini 2.5 Flash |
How It Works
- CEO receives your goal, decomposes it into tasks
- VP Engineering plans engineering tasks and assigns to managers
- Engineering Managers break tasks into subtasks for software engineers
- Software Engineers implement the work
- QA Engineer tests deliverables
- CEO performs final verification
Agent Loop
Each agent follows this cycle:
PLAN → ACT → OBSERVE → REFLECT → VERIFYTools available to agents: create_task, assign_task, escalate, log_decision, save_artifact, final_verify, get_department_status, request_user_clarification
Requirements
- Node.js >= 18
- At least one LLM provider configured (Ollama works out of the box with no API key)
License
MIT
