Agentle4j

The Modern Java Library for OpenAI Responses API

Type-safe, streaming-enabled, production-ready AI integration for Java

Get Started API Reference

---

✨ Key Features

• Agent Framework

  ---

  Complete agent system with tools, guardrails, memory, and handoffs

• Type-safe API

  ---

  Full OpenAI Responses API with Java records

• Real-time Streaming

  ---

  Virtual thread-based streaming with callbacks

• Structured Outputs

  ---

  Strongly-typed JSON responses

• Function Calling

  ---

  Auto-generated JSON schemas from Java classes

• OpenTelemetry

  ---

  Built-in observability with spans/metrics

---

⚡ Performance

Agent Instantiation Benchmark

Measures time to create an agent instance (not API calls). Lower is better.

╔═══════════════════════════════════════════════════════════════════╗
║           AGENT INSTANTIATION BENCHMARK (lower = better)          ║
╚═══════════════════════════════════════════════════════════════════╝

Framework        │  Time (μs) │  Memory (KiB) │ Language
─────────────────┼────────────┼───────────────┼─────────────────────
🚀 Agentle4j     │       0.50 │         0.39  │ Java 21+ (this lib)
AGNO             │       3.00 │         6.60  │ Python
PydanticAI       │     170.00 │        28.71  │ Python
CrewAI           │     210.00 │        65.65  │ Python
LangGraph        │   1,587.00 │       161.43  │ Python
──────────────────────────────────────────────────────────────────────

Important Caveats

• Cross-language comparisons are inherently unfair. Java's JVM provides different performance characteristics than Python's interpreter.
• This benchmark only measures agent instantiation time, not actual LLM inference or end-to-end latency.
• We haven't benchmarked against LangChain4J or Spring AI yet—contributions welcome!
• Real-world performance depends heavily on network latency, model choice, and payload size.

Java Alternatives

Library	Focus	Notes
Agentle4j	Agents-first, OpenAI Responses API	This library
LangChain4J	General-purpose, many integrations	Mature ecosystem
Spring AI	Spring ecosystem integration	Production-ready

---

🚀 Quick Example

import com.paragon.responses.Responder;
import com.paragon.responses.payload.CreateResponsePayload;

// Create a responder
Responder responder = Responder.builder()
    .openRouter()
    .apiKey("your-api-key")
    .build();

// Build your request
var payload = CreateResponsePayload.builder()
    .model("openai/gpt-4o-mini")
    .addDeveloperMessage("You are a helpful assistant.")
    .addUserMessage("Hello!")
    .build();

// Get the response
Response response = responder.respond(payload).join();
System.out.println(response.outputText());

---

📦 Installation

MavenGradle

<dependency>
    <groupId>io.github.paragon-intelligence</groupId>
    <artifactId>agentle4j</artifactId>
    <version>0.1.0</version>
</dependency>

implementation 'io.github.paragon-intelligence:agentle4j:0.1.0'

---

📚 Documentation

Section	Description
Installation	Detailed setup instructions
Getting Started	Quick start guide
Guides	In-depth feature guides
API Reference	Auto-generated Javadoc
Examples	Code samples

---

🤝 Contributing

We welcome contributions! Please see our GitHub repository for more information.

---

Made with ❤️ by Paragon Intelligence