Spice is a portable SQL query, search and LLM-inference engine, written in Rust, that you run as a real-time analytics node next to your operational database. Built by Spice AI, it federates live data from databases, warehouses and lakes, accelerates it locally, and serves it to applications and AI agents through SQL and HTTP APIs. The open-source project has about 3,100 stars on GitHub and is in very active development in 2026.
The runtime connects to sources such as PostgreSQL, MySQL, Databricks, Snowflake, DuckDB and object storage, then materializes “accelerated” copies in engines like DuckDB or Apache Arrow with configurable refresh, so queries that would hammer production systems return in milliseconds. On top of that data layer, Spice adds full-text, vector and hybrid search, embeddings, and model serving: you can run local models or route to hosted LLMs and expose a unified OpenAI-compatible endpoint grounded in your live data. Datasets, models and tools are declared in a YAML “spicepod,” making the whole node reproducible and deployable from laptop to Kubernetes. For AI builders, it collapses the usual glue โ CDC pipelines, a vector database, a feature store and an inference gateway โ into one small binary.
The Spice runtime is free and open source under the Apache-2.0 license. Spice AI also offers the managed Spice Cloud Platform with paid tiers for teams that want hosted infrastructure.
Spice AI is for developers building data-grounded AI apps and agents who need fast, fresh access to operational data โ plus search and model inference โ without assembling a stack of separate services.