Offline Intelligence Library

High-performance LLM inference engine with advanced memory management and context orchestration capabilities. Built in Rust for maximum performance across Windows, macOS, and Linux platforms.

Architecture Overview

This project follows a dual-licensing model:

• Open Source Core (80%): Publicly available under Apache 2.0 license
• Proprietary Extensions (20%): Private plugins for advanced features

Core Components (Public)

• LLM Integration Engine
• Basic Memory Management
• Configuration System
• Metrics and Telemetry
• API Proxy Layer
• Administration Interface

Proprietary Extensions (Private/Future)

• Advanced Context Management (context_engine)
• Key-Value Cache System (cache_management)
• Enhanced Memory Components
• Advanced API Features

Platform Support

Language Bindings

The library provides native bindings for multiple languages:

Native Rust

Direct access to all core functionality:

use offline_intelligence::{Config, run_server};

let config = Config::from_env()?;
run_server(config).await?;

Python

Install via pip:

pip install offline-intelligence

Usage:

from offline_intelligence import Config, run_server

config = Config.from_env()
run_server(config)

C++

CMake integration:

#include <offline_intelligence/offline_intelligence.h>

auto config = offline_intelligence::Config::from_env();
offline_intelligence::run_server(config);

JavaScript/Node.js

NPM package:

npm install offline-intelligence

Usage:

const { Config, runServer } = require('offline-intelligence');

const config = Config.fromEnv();
runServer(config);

Java

Maven dependency:

<dependency>
    <groupId>com.offlineintelligence</groupId>
    <artifactId>offline-intelligence-java</artifactId>
    <version>0.1.0</version>
</dependency>

Usage:

import com.offlineintelligence.Config;
import com.offlineintelligence.Server;

Config config = Config.fromEnv();
Server.runServer(config);

Building from Source

Prerequisites

• Rust 1.70+
• CMake 3.16+ (for C++ bindings)
• Python 3.8+ (for Python bindings)
• Node.js 16+ (for JavaScript bindings)
• Java 11+ (for Java bindings)

Build Process

Windows

build.bat

Linux/macOS

chmod +x build.sh
./build.sh

Build Output

The build process creates distribution packages in the dist/ directory:

• rust/ - Native Rust binaries
• python/ - Python wheels
• cpp-lib/ - C++ libraries and headers
• javascript/ - Node.js packages
• java/ - Java JAR files

Configuration

The library uses environment variables for configuration:

# Core settings
LLAMA_BIN=/path/to/llama-server
MODEL_PATH=/path/to/model.gguf
API_HOST=127.0.0.1
API_PORT=8000

# Resource allocation
THREADS=auto
GPU_LAYERS=auto
CTX_SIZE=auto
BATCH_SIZE=auto

API Endpoints

Core Endpoints

• POST /generate/stream - Stream generation
• GET /healthz - Health check
• GET /readyz - Readiness check
• GET /metrics - Prometheus metrics

Admin Endpoints

• GET /admin/status - System status
• POST /admin/load - Load model
• POST /admin/stop - Stop backend

Memory Endpoints

• GET /memory/stats/{session_id} - Memory statistics
• POST /memory/optimize - Optimize memory
• POST /memory/cleanup - Cleanup memory

Performance Characteristics

• Low Latency: Optimized for real-time inference
• Memory Efficient: Smart caching and garbage collection
• Multi-threaded: Automatic thread pool management
• GPU Accelerated: CUDA support for NVIDIA GPUs

Contributing

We welcome contributions to the open-source core components. Please see our Contributing Guide for details.

License

• Core library: Apache 2.0 License
• Proprietary extensions: Commercial licensing available

Support

For support, please open an issue on our GitHub repository or contact our team at support@offlineintelligence.com.

Roadmap

Short Term (0.2.0)

• Enhanced documentation
• Additional language bindings
• Performance optimizations

Medium Term (0.3.0)

• Plugin architecture for proprietary extensions
• Cloud deployment support
• Enhanced monitoring capabilities

Long Term (1.0.0)

• Full commercial plugin ecosystem
• Enterprise features
• Advanced orchestration capabilities