AgentFlow

AgentFlow is a workflow orchestration engine designed specifically for building and managing AI agent operations. It provides a reliable, strongly-typed environment to string together LLM calls, data fetching, and processing steps into complex, repeatable workflows.

Why AgentFlow?

Building reliable AI applications often means chaining together multiple unpredictable steps. AgentFlow brings the stability of the Java/Spring Boot ecosystem to this problem, offering a structured way to define, execute, and monitor agent-based workflows. It handles the "boring" but critical parts like state management, retries, and failure recovery so you can focus on the agent logic.

Key Features

• YAML-based Workflows: Define complex agent interactions in readable, version-controllable YAML files.
• Built-in Agents: Ready-to-use integrations for HTTP requests, Web Search (DuckDuckGo), and LLM providers (Groq, OpenRouter, OpenAI).
• Sync & Async Execution: Run workflows synchronously or queue them for async processing via Kafka.
• Resilient Execution: Automatic retries, error handling, and persistent state management using PostgreSQL and Redis.
• Scalable: Built on Spring Boot 3 and designed to scale horizontally with Kafka-based message processing.

Getting Started

Prerequisites

• Java 21 or higher
• Maven
• Docker (for supporting services)

Quick Start

1. Clone the repository.
2. Copy .env.example to .env and add your API keys (see SETUP.md).
3. Start the supporting services:

   docker-compose up -d

4. Run the application:

   ./mvnw spring-boot:run

Workflow Example

Here is what a simple workflow looks like in AgentFlow:

name: research-article
version: "1.0"
steps:
  - id: search_topic
    agent: search
    config:
      query: "latest AI developments"
      maxResults: 3
  
  - id: summarize
    agent: llm
    dependsOn: [search_topic]
    config:
      prompt: "Summarize these results: ${steps['search_topic'].outputs.results}"

Execute synchronously or asynchronously:

# Sync execution (waits for completion)
POST /api/workflows/{id}/execute

# Async execution (returns immediately, processes via Kafka)
POST /api/workflows/{id}/execute?async=true

Documentation

For more detailed information on setup, API keys, and testing, see SETUP.md. Architecture details are in the docs/ directory.

Contributing

This project is open for contributions. Please verify your changes before submitting a pull request.

License

Distributed under the MIT License. See LICENSE for more information.