Powering Enterprise-Grade AI Applications & Agents with Operational Data
Spice AI partners with Databricks to help organizations build real-time, data-driven applications and AI agents. Through integration with the Databricks Data Intelligence Platform, Spice AI enables customers to build faster and more reliable AI apps and agents that extend across on-premises, cloud, and edge more rapidly with fewer errors at reduced cost.
With Spice AI + Databricks, Enterprises can:
Build high-performance, secure, and business-ready AI applications with unified access to your Databricks data and models
Query Data For Operational Use Cases
Execute fast, low-latency SQL queries across Databricks, on-premises, and edge sources with Spice.ai's unified engine, enabling real-time applications like inventory tracking or fraud detection.
Embed AI with Applications
Integrate Databricks Mosaic AI model serving and embeddings with the Spice engine to deploy AI features, such as low-latency recommendation systems, search, or predictive maintenance.
Streamline Data Governance
Manage Apache Iceberg and Delta Lake tables using Unity Catalog, enforcing secure access, ensuring compliance, and restricted data access.
Optimize Workload Performance
Use Spice.ai to cache hot data, replicate high-demand datasets, and load-balance hosted AI endpoints, maintaining speed and resilience for applications like real-time dashboards.
Native Support for the Databricks Data Intelligence Platform
Spice AI integrates with the Databricks Data Intelligence platform and features the following capabilities:
Databricks SQL Warehouse and Spark Connect
Integrations for high-performance SQL queries accelerated with DuckDB and SQLite.
Databricks Mosaic AI Integration
Model serving and embeddings integrations to bring MosaicAI alongside applications.
Unity Catalog Support
Comprehensive support for governance and security.
Apache Iceberg & Delta Lake Support
Query and management of open format tables via Unity Catalog.
Service Principal Authentication
M2M & U2M OAuth authentication for enterprise-grade role-based security.
What Our Customers & Databricks Say
Healthcare Analytics and Patient Experience
"Partnering with Spice AI has transformed how NRC Health delivers AI-driven insights. We reduced time-to-market from months to weeks - or even days - and empowered our developers to build dependable, patient-centric tools. Spice isn't just solving our data and AI challenges, it's redefining what's possible in personalized healthcare."
Data Intelligence Platform
"Nearly every enterprise is investing in domain-specific AI agents. Operational data is the key differentiator. We're excited to partner with Spice AI to unlock value on the Databricks Data Intelligence Platform and help customers turn data into intelligence."
FAQs
Common questions about connecting Spice to the Databricks Data Intelligence Platform.
The Spice Databricks connector has three modes: Spark Connect, SQL Warehouse, and Delta Lake. Spark Connect and SQL Warehouse send queries to Databricks compute. Delta Lake mode reads Delta tables directly from Amazon S3, Azure Blob Storage, or Google Cloud Storage. You choose the mode for each dataset with the mode parameter.
The connector accepts a personal access token, machine-to-machine (M2M) OAuth with a service principal, and user-to-machine (U2M) OAuth. M2M OAuth suits production services because it does not depend on an individual user account. Spark Connect mode does not accept U2M OAuth, so use a token or M2M credentials for that mode.
Yes. The Databricks catalog connector registers each Unity Catalog table that matches the include patterns you set. It reads the tables through Spark Connect, SQL Warehouse, or Delta Lake mode. Before you connect, grant read privileges on the catalog to the identity that Spice uses. For M2M OAuth, this identity is the service principal. For a personal access token or U2M OAuth, it is the user. Spark Connect mode also needs Can Attach permission on the cluster.
Spice accelerates hot Databricks tables into a local engine such as DuckDB or SQLite next to the application. Application queries then read local data instead of sending each request to the warehouse. This pattern fits low-latency workloads such as fraud detection, inventory tracking, and real-time dashboards. See data lake acceleration for the architecture.
Yes. Spice calls Mosaic AI Model Serving endpoints as LLMs and as embedding models. You reference an endpoint with the databricks: prefix followed by the endpoint name. Spice can then embed Databricks data for vector and full-text search in the same runtime.
About Spice AI
Spice AI helps enterprises build fast, accurate, and scalable AI applications and agents with its portable, open-source data and AI compute engine. It connects data from disparate sources, simplifies application development, and supports workloads across cloud, edge, and on-premises systems.
Work with Spice AI
Interested in working with Spice AI or looking to learn more about our work? We're always looking for our next big challenge. Take a deeper look at our enterprise offerings.
Visit Spice.ai (opens in a new tab)