In January 2025, Spice announced 1.0 stable, marking the transition from an open-source project to an enterprise-grade, production-ready platform. Spice has shipped 35 stable releases and 11 major releases since then.

From day one, Spice was designed to simplify building modern, intelligent applications. In 2025 that vision turned into reality. Spice now serves as the data and AI substrate for global, production workloads at enterprises like Twilio and Barracuda - where mission-critical applications query, search, and reason over big data in real time.

These data and AI workloads impose fundamentally different demands on the data layer than previous generations of applications. Instead of the complexity of multiple query engines, search platforms, caches, and inference layers, Spice brings this functionality into a single, high-performance data and AI stack. Development teams can query operational databases, data lakes, analytical warehouses, and more with a single SQL interface, while taking advantage of built-in acceleration, hybrid search, and AI.

All of this is delivered by a fully open-source engine built in Rust that can be deployed anywhere - as a sidecar, at the edge, in the cloud, or in enterprise clusters. Developers have complete optionality based on their access patterns and business requirements. 

Below are some of the major features that defined 2025 across the core pillars of the Spice platform: federation and acceleration, search, and embedded LLM inference.

SQL federation and acceleration is at the core of Spice and the applications it enables; enterprise AI applications depend on contextual data drawn from many different systems, and that data must be fast and available to search and reason over in real time.

In 2025, Spice simplified querying across disparate data sources while improving performance, scale, and reliability while the connector ecosystem significantly expanded, enabling teams to ingest and combine data across any source.

Spice leans into the industry shift to object storage as the source of truth for applications. These workloads are often multi-terabyte datasets using open data lake formats like Parquet, Iceberg, or Delta that must serve data and search queries to applications with sub-second performance.

Existing data accelerators like DuckDB, are fast and simple for datasets up to 1TB, however for multi-terabyte workloads, a new class of accelerator is required.

So we built Spice Cayenne, the next-generation data accelerator for high volume and latency-sensitive applications. 
Spice Cayenne combines Vortex, the next-generation columnar file format from the Linux Foundation, with a simple, embedded metadata layer. This separation of concerns ensures that both the storage and metadata layers are fully optimized for what each does best. Cayenne delivers better performance and lower memory consumption than the existing DuckDB, Arrow, SQLite, and PostgreSQL data accelerators.

Spice Founder Luke Kim demonstrated and walked through the details of the Cayenne architecture in a December, 2025 Community Call:

AI applications are only as effective as the data they can retrieve and reason over. Beyond extracting data, they need to search across both structured and unstructured sources to surface the most relevant context at query time. In 2025, search evolved into a core primitive of the Spice platform, designed to operate natively across federated datasets.

In July, Spice introduced native support for Amazon S3 Vectors as a day 1 launch partner at the AWS Summit in NYC. Vector similarity search, structured filters, joins, and aggregations can now be executed in SQL within Spice without duplicating data. 

Developers can make vector searches using SQL or HTTP and combine similarity search with relational predicates and joins. Spice pushes filters down to S3 Vectors to minimize data scanned, delivering scalable sub-second query performance with the flexibility of SQL.

The Spice team presented a live demo of Spice and Amazon S3 Vectors at 2025 AWS re:Invent:

Spice deepened its AI capabilities by making LLM inference via SQL native within the query engine. LLMs can be invoked directly in SQL alongside federated queries, joins, and transformations, helping teams move from raw data to insights all within SQL.

The ai() SQL function enables developers to invoke LLMs directly within SQL for bulk generation, classification, translation, or analysis. Inference runs alongside joins, filters, aggregations, and search results without additional application-layer plumbing. Developers can transform federated data into structured insights extracting data, call external completion APIs, or orchestrate separate pipelines.

Check out a live demo of the AI SQL function here:

2025 was a major year for Spice as it grew from single-node data acceleration to a multi-node data, search, and AI platform. In 2026, Spice 2.0 will focus on bringing multi-node distributed query execution to GA, alongside continued improvements to search, acceleration, and AI primitives. These investments will help deliver even more predictable performance and operational simplicity.

The mission remains the same: to provide a durable, open data substrate that helps teams build and scale the next generation of intelligent, data and AI-driven applications.

Interested in seeing it for yourself? Get started with the open source runtime or cloud platform today.