skardi

High-performance query engine for AI and agents, powered by Apache DataFusion.

Query files, databases, data lakes, and vector stores with SQL — and serve results as parameterized pipelines.

Modules

Installation

[dependencies]
skardi = "0.1"

# With ONNX model inference
skardi = { version = "0.1", features = ["onnx"] }

Quick Start

Execute a SQL query

use skardi::engine::datafusion::DataFusionEngine;
use skardi::engine::Engine;
use datafusion::prelude::*;

#[tokio::main]
async fn main() -> anyhow::Result<()> {
    let ctx = SessionContext::new();
    ctx.register_csv("products", "data/products.csv", CsvReadOptions::new()).await?;

    let engine = DataFusionEngine::new(ctx);
    let result = engine.execute("SELECT * FROM products WHERE price < 100").await?;

    println!("{} rows returned", result.num_rows());
    Ok(())
}

Load and run a pipeline

use skardi::pipeline::pipeline::{Pipeline, StandardPipeline};
use datafusion::prelude::*;
use std::sync::Arc;

#[tokio::main]
async fn main() -> anyhow::Result<()> {
    let ctx = Arc::new(SessionContext::new());

    // Pipeline YAML defines a parameterized SQL query
    let pipeline = StandardPipeline::load_from_file("pipeline.yaml", ctx).await?;

    println!("Pipeline: {} v{}", pipeline.name(), pipeline.version());
    Ok(())
}

A pipeline YAML looks like this:

metadata:
  name: product-search
  version: 1.0.0

query: |
  SELECT name, price FROM products
  WHERE ({brand} IS NULL OR brand = {brand})
    AND price < {max_price}
  LIMIT {limit}

Parameters are automatically inferred from {placeholders} in the SQL.

Feature Flags

Further Reading

• API Reference — full rustdoc for all modules
• GitHub Repository — server, CLI, demos, and deployment guide
• Crate Page

License

Apache-2.0