GitHub - 11divyansh/OxyJen: OxyJen is an open-source Java framework for orchestrating LLM workloads with graph-style execution, context-aware memory, and deterministic retry/fallback. It treats LLMs as native nodes (not helper utilities), allowing developers to build multi-step AI pipelines that integrate cleanly with existing Java code.

OxyJen is the missing deterministic AI Runtime for Java & JVM enterprises.

Deterministic AI Workflow Runtime for the JVM - Build complex AI pipelines with simplicity and power.

What is Oxyjen?

Oxyjen is a graph-based orchestration framework for building AI applications in Java. It provides a clean, extensible architecture for connecting LLMs, data processors, and custom logic into powerful workflows.

Think of it as the plumbing for your AI pipelines, you focus on what each step does, Oxyjen handles the execution flow.

"Why Oxyjen When LangChain4j Exists?"

I get it, this is the first question you're thinking. Let me be completely honest.

The Story

I started building Oxyjen without knowing LangChain4j existed. When I discovered it halfway through, I had a choice:

Abandon the project
Find a way to differentiate

I chose to differentiate. I wanted to learn how OSS works. I wanted to build this in public.

How Oxyjen Will Be Different

LangChain4j is a solid framework focused on feature breadth, lots of integrations, lots of tools. That's great for many use cases.

Oxyjen is taking a different path, focused on developer experience and production readiness

Oxyjen is meant for runtime reliability, your graphs will be self-aware and will make sure to provide less failure, even if a node fails, Oxyjen will learn from it and improve.

Features like, async, project loom, parallel processing, java concurrency will lay down the foundation of fail-safe graph structure for Oxyjen.

I'm not here to compete with Langchain4j, I'm here to create a reliable execution engine for devs.

Why Oxyjen?

Modern AI applications need more than just API calls. They need:

Complex workflows with multiple steps
Type safety to catch errors at compile time
Observability to debug what's happening
Testability to ensure reliability
Extensibility to add custom logic

Oxyjen provides all of this with a simple, intuitive API.

Quick Example

// Build a 3-step text processing pipeline
Graph pipeline = GraphBuilder.named("text-processor")
    .addNode(new UppercaseNode())
    .addNode(new ReverseNode())
    .addNode(new PrefixNode("OUTPUT: "))
    .build();

// Execute with context
NodeContext context = new NodeContext();
Executor executor = new Executor();

String result = executor.run(pipeline, "hello world", context);
System.out.println(result);
// Output: OUTPUT: DLROW OLLEH

That's it! Clean, simple, powerful.

Architecture

Oxyjen is built around four core concepts:

1️Graph - The Pipeline Blueprint

A Graph defines the structure of your pipeline - which nodes run in what order.

public class Graph {
    private final String name;
    private final List<NodePlugin<?, ?>> nodes;
    
    // Add nodes to your pipeline
    public Graph addNode(NodePlugin<?, ?> node);
    
    // Get all nodes in execution order
    public List<NodePlugin<?, ?>> getNodes();
}

Think of it as: Your pipeline's DNA - it knows what needs to happen, but doesn't execute anything.

2️NodePlugin - The Processing Unit

A NodePlugin is a single step in your pipeline. Each node transforms input into output.

public interface NodePlugin<I, O> {
    // Core processing logic
    O process(I input, NodeContext context);
    
    // Unique identifier for this node
    default String getName() { 
        return this.getClass().getSimpleName(); 
    }
    
    // Lifecycle hooks for setup/cleanup
    default void onStart(NodeContext context) {}
    default void onFinish(NodeContext context) {}
    default void onError(Exception e, NodeContext context) {}
}

Think of it as: A Lego brick - small, focused, composable.

Example node:

public class SummarizerNode implements NodePlugin<String, String> {
    @Override
    public String process(String input, NodeContext context) {
        context.getLogger().info("Summarizing text...");
        // Your logic here (will be LLM call in v0.2)
        return "Summary: " + input.substring(0, 100);
    }
    
    @Override
    public void onStart(NodeContext context) {
        context.getLogger().info("Summarizer node starting");
    }
}

3️Executor - The Runtime Engine

The Executor runs your graph, calling each node in sequence and passing outputs to inputs.

public class Executor {
    public <I, O> O run(Graph graph, I input, NodeContext context) {
        // Validates graph structure
        // Executes nodes sequentially
        // Handles errors and lifecycle hooks
        // Returns final output
    }
}

Think of it as: The conductor of an orchestra - coordinates everything.

How it works:

Takes your Graph and initial input
For each node:
- Calls onStart() lifecycle hook
- Executes process() with current data
- Calls onFinish() lifecycle hook
- Passes output to next node
Returns final result

4️NodeContext - Shared Memory & State

The NodeContext is shared across all nodes, providing logging and state management.

public class NodeContext {
    // Store/retrieve shared data
    public void set(String key, Object value);
    public <T> T get(String key);
    
    // Logging
    public Logger getLogger();
    public OxyLogger getOxyjenLogger();
    
    // Metadata (e.g., graph name, execution ID)
    public void setMetadata(String key, Object value);
    public <T> T getMetadata(String key);
    
    // Error handling
    public ExceptionHandler getExceptionHandler();
}

Think of it as: A shared notebook that all nodes can read/write to.

Example usage:

public String process(String input, NodeContext ctx) {
    // Log what's happening
    ctx.getLogger().info("Processing: " + input);
    
    // Store intermediate results
    ctx.set("word_count", input.split(" ").length);
    
    // Share data between nodes
    String previousResult = ctx.get("previous_output");
    
    return processedOutput;
}

Complete Working Example

package examples;

import io.oxyjen.core.*;

public class ContentPipeline {
    
    public static void main(String[] args) {
        // Step 1: Define your nodes
        NodePlugin<String, String> validator = new ValidationNode();
        NodePlugin<String, String> processor = new ProcessingNode();
        NodePlugin<String, String> formatter = new FormatterNode();
        
        // Step 2: Build your graph
        Graph pipeline = GraphBuilder.named("content-pipeline")
            .addNode(validator)
            .addNode(processor)
            .addNode(formatter)
            .build();
        
        // Step 3: Create execution context
        NodeContext context = new NodeContext();
        context.set("max_length", 100);
        
        // Step 4: Execute
        Executor executor = new Executor();
        String result = executor.run(pipeline, "Raw input text", context);
        
        System.out.println("Final output: " + result);
        System.out.println("Word count: " + context.get("word_count"));
    }
}

// Example node implementations
class ValidationNode implements NodePlugin<String, String> {
    @Override
    public String process(String input, NodeContext ctx) {
        if (input == null || input.isEmpty()) {
            throw new IllegalArgumentException("Input cannot be empty");
        }
        ctx.getLogger().info("Input validated");
        return input;
    }
}

class ProcessingNode implements NodePlugin<String, String> {
    @Override
    public String process(String input, NodeContext ctx) {
        String processed = input.toUpperCase().trim();
        ctx.set("word_count", processed.split(" ").length);
        ctx.getLogger().info("Text processed");
        return processed;
    }
}

class FormatterNode implements NodePlugin<String, String> {
    @Override
    public String process(String input, NodeContext ctx) {
        Integer maxLength = ctx.get("max_length");
        String formatted = input.length() > maxLength 
            ? input.substring(0, maxLength) + "..." 
            : input;
        ctx.getLogger().info("Text formatted");
        return formatted;
    }
}

My Vision for Oxyjen

Vision

Bring AI orchestration (LangChain/LangGraph style) to Java.
Build enterprise-first modules: LLM agents, Audit tools, Secure complex Workflow Engine.
Focus on performance, security, and observability.
I'm building this to learn java in a much deeper way.

Phase 6 in progress

RAG support - Vector databases, embeddings, document loaders
Cost management - Budgets, limits, usage tracking
Enterprise features - Audit logs, RBAC, compliance
Multi-tenancy - Isolate data between users/orgs
Circuit breakers - Fail fast when services are down
Streaming responses
Token counting & cost tracking
Async/reactive API

Documentation

Installation

Maven

Add JitPack repository:

<repositories>
  <repository>
    <id>jitpack.io</id>
    <url>https://jitpack.io</url>
  </repository>
</repositories>

Add dependency:

<dependency>
  <groupId>com.github.11divyansh</groupId>
  <artifactId>Oxyjen</artifactId>
  <version>v0.4.0</version>
</dependency>

Gradle

repositories {
  maven { url 'https://jitpack.io' }
}

dependencies {
  implementation 'com.github.11divyansh:Oxyjen:v0.4.0'
}

Build from Source

git clone https://github.com/11divyansh/OxyJen.git
cd OxyJen
mvn clean install

After installation, verify by importing:

import io.oxyjen.core.*;
import io.oxyjen.llm.*;
import io.oxyjen.tools.*;
import io.oxyjen.graph.*;

About

Built with ❤️ by Divyansh Bhatt - a BTech CS student who believes Java deserves world-class AI tooling.

This started as a learning project, but I'm committed to making it production-ready. I know this is not big yet, but lets make it valuable.

Get Involved

Star this repo to follow the journey and be a part of it
Report bugs via Issues
Suggest features via Discussions
Contribute code or documentation
Share on Twitter/LinkedIn if you find it useful

** Watch for updates on v0.5 progress!**

License

Apache 2.0 (open-source, enterprise-friendly)

推荐订阅源

Hacker News - Newest: "AI"