Spice AI builds Spice, an open-source engine for enterprise data that combines SQL querying, search, and LLM inference in one. Applications and AI agents query databases, data lakes, and data warehouses through a single SQL endpoint. Spice can also maintain an analytical copy of operational databases, so analytical reads do not load production systems. Teams can run Spice on their own infrastructure or use the managed Spice Cloud Platform.
Key capabilities:
- Fast, federated access to all data: Connect to and query operational databases, data lakes, and data warehouses across the enterprise. Materialize and accelerate working sets into memory or disk for millisecond-level access.
- SQL federation and acceleration: Achieve sub-second query performance across sources, with official claims of up to 100x faster queries, up to 80% lower lakehouse spend, and 2x better data reliability.
- Real-time analytics without affecting production: Deploy analytical copies of operational data in minutes for fast, isolated real-time queries.
- Hybrid SQL search: Run hybrid search using standard SQL, combining structured filtering, vector similarity, and keyword matching into a single query, and rank the combined results by relevance.
- Call LLMs within the query layer: Call hosted or local LLMs through SQL UDFs or inline natural language, completing text translation, summarization, entity classification, and query result enrichment within the Spice runtime.
- Isolated data access for AI agents: Provide applications and agents with isolated, least-privilege datasets without direct database access, supporting RAG, agents, and AI workflows.
- Distributed observability: Get end-to-end tracing for SQL, embeddings, search, and LLM calls, making it easy to debug, measure latency, and prove ROI.
How it is used: Spice is written in Rust, uses Apache DataFusion as its SQL query engine, and data moves through the runtime in the Apache Arrow columnar format. Apache Ballista runs distributed queries on a cluster. The Spice Cayenne accelerator stores local copies in the Vortex columnar format. AI agents can query Spice through three interfaces: the SQL API, an OpenAI-compatible chat API, and an MCP server.
How it is deployed: Spice can run on a laptop, on-premises, at the edge, or on any cloud. Teams can deploy it as a sidecar next to an application, a sharedmicroservice, or a distributed cluster on Kubernetes. Spice Cloud Platform runs Spice as a managed service.
Use cases: From messaging platforms to security systems, the company says companies such as Twilio and Barracuda rely on Spice to deliver low-latency applications and AI agents at scale. Twilio's software architect said Spice opened the door to moving critical control plane datasets into the runtime path next to services. Barracuda's principal software engineer said it works as soon as it starts, with astonishing responsiveness. NRC Health's vice president of technology said that working with Spice AI changed the way it delivers AI-driven insights, shortening time to market from months to weeks or even days.
Limitations and notes:
- Spice does not replace operational databases. It runs alongside existing databases, data lakes, and data warehouses, which remain the source of truth. Applications continue to write to operational databases, while Spice provides read services through SQL federation and acceleration.
- The open-source version of Spice.ai is free to use under the Apache 2.0 license, with community support available on GitHub and Slack. Spice Cloud offers a managed service with subscription and usage-based plans. Enterprise plans add commercial licensing, 24/7 support, and SLAs.
- Specific pricing details are not provided at this time.
- Installation commands, code examples, and similar details are not provided in the materials and are not available at this time.
Spice open-source documentation and Cookbook provide more than 80 guides and examples to help build data-driven AI applications and agents.
