active_agent 🤖

The Native Multi-Agent AI Framework for Ruby & Rails.

Inspired by Python's CrewAI and Microsoft AutoGen, active_agent provides a clean, idiomatic Ruby DSL to orchestrate teams of autonomous AI agents with roles, backstories, tools, tasks, and multi-provider LLMs.

---

Key Features 🚀

• 👥 Autonomous Agent Teams: Define specialized personas (Agent) with goals, backstories, and assigned capabilities.
• 🛠️ Custom Executable Tools: Agents dynamically select and execute custom Ruby tools (ActiveAgent::Tool) to perform real-world tasks (Web Search, Database Queries, API calls).
• 📋 Task Orchestration: Assign discrete objectives (Task) with expected outputs and context passing across workflow steps.
• 🌐 Multi-Provider LLM Adapters: Built-in support for OpenAI (gpt-4o), Anthropic Claude (claude-3-5-sonnet), Google Gemini (gemini-1.5-flash), and Ollama (Local AI).
• ⚡ Sequential & Parallel Processes: Run agent teams in sequential pipelines where outputs pass from step to step.

---

Installation

Add this line to your application's Gemfile:

gem 'active_agent_ai'

And then execute:

bundle install

---

Quickstart Example

Define tools, agents, tasks, and kick off your agent team in 100% native Ruby:

require 'active_agent'

# 1. Define a Custom Executable Tool
class WebSearchTool < ActiveAgent::Tool
  description "Searches the web for latest software benchmarks and language features"
  param :query, type: :string, desc: "Search query", required: true

  def perform(query:)
    # Execute actual search or API call here
    "Ruby 3.4 introduces Prism parser by default and 20% YJIT performance gains."
  end
end

# 2. Define Specialized AI Agents
researcher = ActiveAgent::Agent.new(
  role: "Senior Tech Researcher",
  goal: "Find cutting-edge developments in Ruby & Rails",
  backstory: "An expert language analyst with 10 years of compiler and AST experience.",
  tools: [WebSearchTool],
  provider: :openai # or :claude, :gemini, :ollama
)

writer = ActiveAgent::Agent.new(
  role: "Technical Journalist",
  goal: "Draft engaging tech articles from technical research reports",
  backstory: "A skilled writer capable of explaining complex software benchmarks simply.",
  provider: :openai
)

# 3. Define Tasks
research_task = ActiveAgent::Task.new(
  description: "Research Ruby 3.4 JIT benchmarks and Prism parser updates",
  expected_output: "Detailed technical report with benchmarks",
  agent: researcher
)

write_task = ActiveAgent::Task.new(
  description: "Write a high-converting blog post based on the research findings",
  expected_output: "Formatted Markdown article ready for publication",
  agent: writer
)

# 4. Assemble and Kickoff Team
team = ActiveAgent::Team.new(
  agents: [researcher, writer],
  tasks: [research_task, write_task],
  process: :sequential
)

result = team.kickoff
puts result

---

Multi-Provider Support

active_agent seamlessly works with all top LLM providers:

# OpenAI
agent = ActiveAgent::Agent.new(role: "Coder", goal: "Refactor", provider: :openai, model: "gpt-4o")

# Anthropic Claude
agent = ActiveAgent::Agent.new(role: "Analyst", goal: "Audit", provider: :claude, model: "claude-3-5-sonnet-20241022")

# Google Gemini
agent = ActiveAgent::Agent.new(role: "Data Engine", goal: "Parse", provider: :gemini, model: "gemini-1.5-flash")

# Ollama (100% Local AI - No API keys needed)
agent = ActiveAgent::Agent.new(role: "Local Assistant", goal: "Summarize", provider: :ollama, model: "llama3")

---

License

MIT License.