spiceai
Add a real-time analytics node to your operational database. Spice is a portable, accelerated SQL query, search, and LLM-inference engine in Rust for data-groun
What is spiceai?
Add a real-time analytics node to your operational database. Spice is a portable, accelerated SQL query, search, and LLM-inference engine in Rust for data-grounded AI apps and agents.
Overview of spiceai
What it is
Spice (Spice.ai Open Source) is a data and AI engine that brings data to applications. It connects to databases, data warehouses, data lakes, and APIs, and serves data through standard APIs with SQL query, search, and LLM inference built in. Queries on accelerated datasets return in milliseconds because Spice keeps a local, continuously refreshed copy. Other datasets are queried in place with query push-down. Spice is a single binary written in Rust on Apache DataFusion, Apache Arrow, and Vortex. It can run as a sidecar, standalone service, or multi-node distributed cluster. The repository contains the Apache-2.0 runtime (spiced) and CLI (spice). Spice.ai Cloud and Spice.ai Enterprise are optional managed and enterprise distributions built on it.
Key features
- Real-time analytics node for operational databases: native CDC replication (WAL, binlog, change streams) for PostgreSQL, MySQL, and MongoDB with ~2-second freshness and zero analytical load on production.
- Millisecond queries against a sandboxed working set, microsecond repeat queries from the results cache, and transparent delegation to a distributed cluster.
- Distributed query with Apache Ballista: multi-active schedulers coordinated through object storage. 2.9x faster than single-node DataFusion on TPC-H SF100, 8x less RAM than Spark.
- Spice Cayenne acceleration engine: high-rate CDC ingestion and low-latency queries on the same table. 1.5x faster than DuckDB with 3x less memory on TPC-H SF100, 26x faster than Spice 1.x on TPC-DS SF100, 100x faster random access vs. Parquet.
- Petabyte-scale hybrid search: native Amazon S3 Vectors, Tantivy BM25, DuckDB HNSW, and Elasticsearch kNN with reciprocal rank fusion and reranker UDTFs in a single SQL query.
- AI-native runtime: OpenAI-compatible APIs, MCP server + gateway, LLM memory, NSQL text-to-SQL, multi-vector ColBERT-style embeddings, provider-aware prompt caching.
- 30+ data connectors with advanced query push-down: Postgres, MySQL, Snowflake, Databricks, Iceberg, Delta Lake, S3, Spark, MSSQL, DynamoDB, MongoDB, GitHub, SharePoint, Kafka, and more.
Use cases
- Build data-grounded AI apps and agents in minutes without pipelines or glue code.
- Add a real-time analytics node to operational databases (PostgreSQL, MySQL, MongoDB) to offload analytical queries from production.
- Federate queries across multiple data sources (databases, data lakes, APIs) with a single SQL interface.
- Serve LLM inference and search (vector, full-text, hybrid) alongside structured data queries.
- Use as a sidecar for localhost-latency data access in microservices or as a distributed cluster for large-scale workloads.
Getting started
See the repository README for installation and quickstart instructions.
License and pricing
Licensed under Apache-2.0. Open source (free).
Based on the project's public README and repository metadata.