Spice is a Rust-based data query and AI compute engine designed for data-grounded AI applications and agents. It provides a unified runtime for SQL querying across diverse data sources and serving AI models, simplifying development for AI-native applications by co-locating data and compute.

How It Works

Spice combines Apache DataFusion and DuckDB for OLAP workloads and SQLite for OLTP, offering dual-engine acceleration. It supports disaggregated storage, allowing local materialization of data subsets with applications. The engine exposes multiple standard APIs, including SQL (Arrow Flight SQL, ODBC, JDBC, ADBC), OpenAI-compatible HTTP APIs for LLM inference, and Iceberg Catalog REST APIs, enabling seamless integration with existing tools and workflows.

Quick Start & Requirements

• Install: CLI via curl or brew (macOS), PowerShell (Windows). Docker image available.
• Prerequisites: None explicitly stated for basic CLI installation. GPU/Metal acceleration for AI models is supported but not required for core functionality.
• Setup: Local quickstart involves spice init, spice run, spice add, and spice sql.
• Docs: 📄 Docs, ⚡️ Quickstart, 🧑‍cookbook

Highlighted Details

• AI-Native Runtime: Integrates data querying and AI inference in a single engine.
• Application-Focused Deployment: Designed for distributed, per-application or per-tenant instances.
• Data Federation: SQL querying across databases, data lakes, and warehouses.
• OpenAI Compatibility: Serves local models and acts as a gateway for hosted models.

Maintenance & Community

• Active development with a public roadmap.
• Community channels: Discord, X, LinkedIn.
• Contributions welcome via CONTRIBUTING.md.

Licensing & Compatibility

• Apache 2.0 License. Permissive for commercial use and closed-source linking.

Limitations & Caveats

• Some data connectors and model providers are in Alpha or Beta stages, indicating potential instability or incomplete functionality.