airust

🧠 airust is a modular, trainable AI library in Rust.
It supports compile-time knowledge through JSON files and enables simple prediction engines for natural language inputs.

🚀 Features

• 🧩 Modular architecture with the TrainableAgent trait
• 🧠 Multiple built-in agents:
  • SimpleAgent (exact matching)
  • FuzzyAgent (Levenshtein similarity)
  • ContextAgent (considers conversation context)
  • TfidfAgent (uses BM25 algorithm for better similarity matching)
  • StructuredAgent (supports structured response formats)
• 💾 Compile-time knowledge via knowledge/train.json
• ⚖️ Weighted training data for more precise answers
• 📋 Extensible knowledge base at runtime
• 🔍 Advanced text recognition with TF-IDF and BM25
• 🏷️ Support for structured responses (Text, Markdown, JSON)
• 📦 Easy integration into other projects
• 🖥️ CLI test program included

🔧 Usage

In your project

[dependencies]
airust = { path = "../airust" }

Example code

use airust::simple_agent::SimpleAgent;
use airust::knowledge::TRAINING_DATA;
use airust::agent::TrainableAgent;

fn main() {
    let mut ai = SimpleAgent::new();
    ai.train(&TRAINING_DATA);
    let answer = ai.predict("What is airust?");
    println!("Answer: {}", answer);
}

📂 Training Data

The file knowledge/train.json now also supports weights:

[
  {
    "input": "What is GEL?",
    "output": "A lightweight version control system.",
    "weight": 1.0
  },
  {
    "input": "What is airust?",
    "output": "A modular AI agent in Rust.",
    "weight": 2.0
  }
]

This file is automatically embedded in the binary at build time (build.rs takes care of this).

🖥️ CLI Usage

# Testing different agents
cargo run --bin cli -- simple "What is GEL?"
cargo run --bin cli -- fuzzy "What is Gel"
cargo run --bin cli -- tfidf "Explain airust to me"
cargo run --bin cli -- context "Follow-up question on the topic"

🧪 Testing the Extended Features

Context Agent Testing

# Start the interactive context test
cargo run --bin context_test

The Context Agent stores previous questions and answers to deliver better results in connected conversations.

Dynamic Knowledge Base

# Test the dynamic knowledge database
cargo run --bin knowledge_test

With the dynamic knowledge base, you can at runtime:

• Add new training data
• Save and load the knowledge base
• Make changes to training data

Structured Responses

The StructuredAgent supports different response formats:

• Simple text
• Markdown formatted text
• JSON structured data

# Test structured responses
cargo run --bin structured_test

📊 Advanced Usage

BM25 Algorithm for Better Match Rates

The TfidfAgent uses the BM25 algorithm, an extension of the TF-IDF method, to better recognize semantic similarity between questions:

use airust::tfidf_agent::TfidfAgent;
use airust::knowledge::TRAINING_DATA;
use airust::agent::TrainableAgent;

fn main() {
    let mut ai = TfidfAgent::new();
    ai.train(&TRAINING_DATA);
    // Finds answers even with differently phrased questions
    let answer = ai.predict("Explain to me what airust can do");
    println!("{}", answer);
}

📃 License

MIT

Developed with ❤️ in Rust.
This crate is open for contributions and extensions.