llms.txt
Compare with other sites' llms.txt files →# Spice AI
> Spice.ai is a data and AI platform that combines federated SQL query, hybrid search, and LLM inference in a portable, open-source runtime
Spice AI provides a unified data and AI infrastructure platform. Our open-source runtime enables federated SQL query across multiple data sources, hybrid vector and full-text search, and seamless LLM inference integration.
## Documentation
- [Spice.ai Platform Documentation](https://spice.ai/docs): Documentation for Spice Cloud, Spice Enterprise, and the Spice.ai platform API
- [Spice.ai OSS Documentation](https://spiceai.org/docs): Documentation for the open-source Spice runtime
- [GitHub Repository](https://github.com/spiceai/spiceai): Open-source codebase and examples
## Platform
- [Analytics Replica](https://spice.ai/platform/analytics): Add a real-time analytics replica to PostgreSQL, MySQL, MongoDB, and DynamoDB. Spice streams each change from the native change log into a local accelerated copy. Analytical queries run on the replica, with no ETL.
- [LLM Inference](https://spice.ai/platform/llm-inference): Call LLMs from SQL with ai(), get typed answers with decision functions such as ai_if and ai_classify, or ask a question and get SQL back.
- [RAG & Agentic Retrieval](https://spice.ai/platform/retrieval-augmented-generation): Ground RAG pipelines and AI agents in enterprise data. Spice combines hybrid search, reranking, SQL federation, and model inference in one runtime.
- [SQL Federation & Acceleration](https://spice.ai/platform/sql-federation-acceleration): Spice joins data from many sources in one SQL query, and keeps hot tables in a local engine for fast reads. No ETL pipeline is necessary.
## Use Cases
- [Agent Data Serving](https://spice.ai/use-case/agent-data-serving): Spice gives agents fast, secure access to operational databases, warehouses, and APIs. Run analytical queries on an isolated replica of your operational data, kept current over high-throughput CDC.
- [Analytics on Operational Data](https://spice.ai/use-case/analytics): Spice replicates operational databases into a columnar analytics replica, so agents and dashboards run large analytical queries on real-time operational data.
- [Datalake Accelerator](https://spice.ai/use-case/datalake-accelerator): Accelerate query performance in your data lake with Spice. Run SQL locally on federated datasets for up to 100x faster performance.
- [Query Across Sources Without ETL](https://spice.ai/use-case/query-across-sources): Spice federates SQL across operational databases, warehouses, data lakes, and APIs, and accelerates hot data next to the app, with no ETL pipeline to maintain.
## Features
- [AI Model Serving](https://spice.ai/feature/ai-model-serving): Serve hosted and local LLMs from the Spice runtime. Call them through an OpenAI-compatible API or the ai() SQL function, grounded in your data.
- [Distributed Query](https://spice.ai/feature/distributed-query): Distributed query in Spice runs one SQL query across a cluster of executor nodes, coordinated by a scheduler built on Apache Ballista and DataFusion, for data lake scans too large for one node.
- [Edge to Cloud Deployments](https://spice.ai/feature/edge-to-cloud-deployments): Deploy Spice anywhere, from lightweight sidecars to enterprise clusters. Choose the architecture that fits your performance, scale, and governance needs.
- [MCP Server & Gateway](https://spice.ai/feature/mcp-server-gateway): Spice is an MCP gateway that serves upstream MCP servers and its own SQL and search tools through one authenticated /v1/mcp endpoint, with per-model tool access.
- [Real-Time Change Data Capture](https://spice.ai/feature/real-time-change-data-capture): Change data capture (CDC) in Spice applies inserts, updates, and deletes from source databases to accelerated datasets in near real time, without full-table refreshes.
- [Secure AI Sandboxing](https://spice.ai/feature/secure-ai-sandboxing): Isolate AI agents from production data. Spice gives each agent a read-only runtime with filtered datasets, scoped tools, API keys, and query tracing.
## Industries
- [Cybersecurity](https://spice.ai/industry/cybersecurity): Spice is a SQL query, search, and AI runtime for cybersecurity applications. It federates and accelerates logs, telemetry, and threat data for detection and triage.
- [Financial Services](https://spice.ai/industry/financial-services): Spice is a SQL query, search, and AI runtime for financial services. It federates and accelerates governed data for fraud detection, search, and traceable AI.
- [SaaS](https://spice.ai/industry/saas): Spice is a SQL query, search, and AI runtime for SaaS applications. It serves live tenant data at millisecond latency and adds LLM features in one runtime.
## Partners
- [Partners](https://spice.ai/partners): Partner with Spice AI to deliver faster data, search, and AI solutions for your customers. Build integrations, co-market solutions, and power data-intensive applications and AI agents together.
- [Spice AI for AWS](https://spice.ai/partners/aws): Build Fast, Scalable AI Applications with Spice AI and Amazon Web Services
- [Spice AI for Databricks](https://spice.ai/partners/databricks): Build Fast, Accurate AI Applications with Spice AI and Databricks
- [Spice AI for NetApp](https://spice.ai/partners/netapp): Build Accelerated, Data-Grounded AI Applications with Spice AI and NetApp ONTAP
## Learn
- [Learn Data & AI](https://spice.ai/learn): Learn about the core technologies behind Spice.ai: SQL federation, data virtualization, RAG, hybrid search, change data capture, the Model Context Protocol, and more.
- [Apache DataFusion vs DuckDB: How to Choose](https://spice.ai/learn/apache-datafusion-vs-duckdb): Compare Apache DataFusion and DuckDB across architecture, extensions, persistence, concurrency, and deployment. Use equivalent SQL examples and workload criteria to choose an engine.
- [Apache Iceberg vs Delta Lake: How to Choose](https://spice.ai/learn/apache-iceberg-vs-delta-lake): Apache Iceberg and Delta Lake are both open table formats that bring transactions, schema evolution, and time travel to data lakes. Learn the key architectural differences and when to use each.
- [Best Alternatives to ETL Pipelines for AI Agents](https://spice.ai/learn/best-alternatives-to-etl-pipelines-for-ai-agents): Objective guide to alternatives to ETL pipelines for AI agents, including federation, CDC, event streaming, and hybrid data architecture patterns.
- [Best Query Engines for Real-Time AI Analytics](https://spice.ai/learn/best-query-engines-for-real-time-ai-analytics): Objective comparison of query engines for real-time AI analytics, covering DataFusion, DuckDB, ClickHouse, Trino, Druid, Pinot, and Spark against AI workload requirements.
- [Best Snowflake Alternatives for Real-Time SQL](https://spice.ai/learn/snowflake-alternatives-for-real-time-sql): Objective guide to Snowflake alternatives for real-time SQL workloads, comparing federation engines, real-time OLAP databases, lakehouse engines, and columnar replicas.
- [Cache Invalidation at Scale: Why Manual Strategies Break](https://spice.ai/learn/cache-invalidation-at-scale): Learn why TTLs, purge-on-write, and event-driven cache invalidation break down as systems scale, and how declarative refresh policies replace hand-written invalidation logic.
- [Caching vs Data Acceleration: How to Choose](https://spice.ai/learn/caching-vs-data-acceleration): Compare caching and data acceleration: two approaches to making data access faster. Learn the key differences in invalidation, freshness, query flexibility, and operational cost, and when each approach fits.
- [Comparing Data Federation Tools for AI Agents: How to Choose](https://spice.ai/learn/comparing-data-federation-tools-for-ai-agents): Compare data federation tool categories for AI agents across latency, freshness, governance, and operational overhead. Learn which approach fits your architecture.
- [Data Lakehouse vs Data Warehouse: How to Choose](https://spice.ai/learn/data-lakehouse-vs-data-warehouse): Compare data lakehouses and data warehouses: two architectures for analytical data. Learn the key differences in storage formats, cost, workload fit, and application-serving latency, and when each is appropriate.
- [Data Virtualization vs Data Replication: How to Choose](https://spice.ai/learn/data-virtualization-vs-replication): Compare data virtualization and data replication: two foundational approaches to data integration. Learn the key differences, tradeoffs, and when each approach is appropriate for your workloads.
- [Everything You Need to Know About AI Agent Data Access](https://spice.ai/learn/ai-agent-data-access): Complete guide to AI agent data access, covering connectivity, low-latency serving, governance, isolation, and how to choose an access pattern.
- [Full-Text Search vs Vector Search: How to Choose](https://spice.ai/learn/full-text-search-vs-vector-search): Full-text search matches exact keywords using BM25 scoring, while vector search finds semantically similar content using embeddings. Learn the key differences, when to use each approach, and how hybrid search combines both for optimal results.
- [How Do You Keep Embeddings in Sync When the Source Data Changes?](https://spice.ai/learn/how-to-keep-embeddings-in-sync): Embeddings go stale when source data changes. Learn how to detect inserts, updates, and deletes, re-embed only the chunks that changed, and keep vector and keyword indexes consistent.
- [How to Build a Resilient Data Layer for AI Agents](https://spice.ai/learn/resilient-data-layer-ai-agents): Learn how to build a resilient, governed data layer for AI agents that query live OLTP and lakehouse data with reliable freshness and isolated read capacity.
- [How to Connect AI Agents to Live Operational Data Without ETL](https://spice.ai/learn/how-to-connect-ai-agents-to-live-operational-data-without-etl): Practical guide to connecting AI agents to live operational data using federation, acceleration, and policy controls instead of batch ETL pipelines.
- [How to Do SQL Query Federation](https://spice.ai/learn/sql-federation): SQL query federation lets you query multiple databases with a single SQL statement without moving data. Learn how to set up federated queries, optimize with predicate pushdown, and combine federation with acceleration.
- [How to Give Each AI Agent Its Own Isolated Data Environment](https://spice.ai/learn/how-to-give-each-ai-agent-its-own-isolated-data-environment): Practical architecture guide for isolating AI agent data environments using scoped credentials, runtime boundaries, and policy enforcement patterns.
- [How to Implement Change Data Capture (CDC)](https://spice.ai/learn/change-data-capture): Change data capture (CDC) tracks row-level changes in databases and streams them in real time. Learn how to implement CDC with log-based, trigger-based, and polling patterns for real-time data pipelines.
- [How to Migrate Data and Search Infrastructure Across Clouds](https://spice.ai/learn/how-to-migrate-data-and-search-infrastructure-across-clouds): Learn how to move databases, data APIs, and search systems between on-premises environments and clouds behind a stable query endpoint, with staged cutover and validation.
- [How to Reduce Data Lakehouse Costs for Agentic Workloads](https://spice.ai/learn/how-to-reduce-data-lakehouse-costs-for-agentic-workloads): Practical framework for reducing data lakehouse costs in agentic workloads by separating serving paths, minimizing expensive query patterns, and optimizing retrieval architecture.
- [How to Sandbox Data Access for AI Agents](https://spice.ai/learn/how-to-sandbox-data-access-for-ai-agents): Step-by-step guide to sandboxing AI agent data access with least-privilege policies, query guardrails, redaction, and runtime controls for production safety.
- [How to Scale Data Infrastructure from a Few Agents to Many](https://spice.ai/learn/how-to-scale-data-infrastructure-from-a-few-agents-to-many): Learn how to scale data infrastructure from a few AI agents to hundreds, covering query fan-out, connection limits, capacity math, and shared data plane patterns.
- [How to Securely Connect AI Agents to Multiple Databases](https://spice.ai/learn/how-to-connect-ai-agents-to-multiple-databases): Learn how to securely connect AI agents to multiple databases using SQL query federation, unified policy controls, local acceleration, and MCP gateways.
- [How to Simplify Data and AI Application Architectures](https://spice.ai/learn/how-to-simplify-data-and-ai-application-architectures): Learn how to collapse complex ETL pipelines, CDC buses, vector database sync, and caching layers into a unified data runtime for modern AI applications.
- [How to Use Text-to-SQL](https://spice.ai/learn/text-to-sql): Text-to-SQL uses large language models to translate natural language questions into SQL queries. Learn how to implement text-to-SQL with schema-aware generation, prompt engineering, and production safeguards.
- [Managed Apache DataFusion: Federated SQL at Scale](https://spice.ai/learn/managed-apache-datafusion): Managed Apache DataFusion provides federated SQL across heterogeneous systems with centralized operations, query optimization, and production controls. Learn how it works and when to use it.
- [Open-Source AI Data Platforms: Licensing, Governance, and Open Core](https://spice.ai/learn/open-source-ai-data-platforms): Guide to evaluating open-source AI data platforms by license type, governance model, and where the open core boundary sits, rather than by the open-source label alone.
- [RAG vs Fine-Tuning: How to Choose](https://spice.ai/learn/rag-vs-fine-tuning): RAG retrieves external data at inference time while fine-tuning embeds knowledge into model weights. Learn the key differences, tradeoffs, and when to use each approach for production AI applications.
- [Real-Time Analytics Guide for AI Agents](https://spice.ai/learn/real-time-analytics-guide-for-ai-agents): Learn how to deliver real-time analytics for AI agents using analytics replicas, change data capture, low-latency serving, and governed access without ETL.
- [Sidecar vs Microservice Architecture: How to Choose](https://spice.ai/learn/sidecar-vs-microservice-architecture): Sidecar and microservice are two deployment architectures for data and AI runtimes. Learn the key differences in latency, scaling, resource usage, and when to use each pattern.
- [SQL Federation vs ETL: How to Choose](https://spice.ai/learn/sql-federation-vs-etl): SQL federation and ETL are two approaches to accessing data across distributed systems. Learn the key differences, when to use each, and how modern platforms combine both for real-time performance.
- [What Are Analytics Replicas for Operational Data?](https://spice.ai/learn/analytics-replica-operational-data): Learn how analytics replicas run analytical queries on operational data without sending unbounded read load to production databases.
- [What are Embeddings?](https://spice.ai/learn/embeddings): Embeddings are dense vector representations of text, images, or code that capture semantic meaning. Learn how embedding models work, what dimensions represent, and how embeddings enable semantic search, RAG, and classification.
- [What Are Management, Control, and Data Planes?](https://spice.ai/learn/management-control-and-data-planes): The management, control, and data planes divide a data or AI platform into the parts that govern it, change it, and serve requests. Learn what each plane does.
- [What is a Data Substrate?](https://spice.ai/learn/data-substrate): A data substrate is a co-located data and AI layer that provides applications with sub-millisecond access to any data source. Learn how it differs from a data warehouse, lake, or lakehouse, and why it is an architectural primitive rather than a destination.
- [What is a Hybrid Data Architecture?](https://spice.ai/learn/hybrid-data-architecture): A hybrid data architecture combines application sidecars for sub-millisecond reads with a centralized cluster for data ingestion, acceleration, and distributed compute. Learn how the sidecar-cluster pattern works and when to use it.
- [What is Apache Arrow?](https://spice.ai/learn/apache-arrow): Apache Arrow is a cross-language development platform for in-memory columnar data. Learn how Arrow works, its columnar memory format, Arrow Flight, Arrow IPC, and how it compares to Parquet, Protocol Buffers, and CSV.
- [What is Apache Ballista?](https://spice.ai/learn/apache-ballista): Apache Ballista is a distributed SQL query engine that extends Apache DataFusion across multiple nodes. Learn how Ballista works, its architecture, and how it compares to Spark and Trino.
- [What is Apache DataFusion?](https://spice.ai/learn/apache-datafusion): Apache DataFusion is an open-source, extensible SQL query engine written in Rust. Learn how DataFusion works, its architecture, how it compares to Trino and DuckDB, and how teams extend it for production use cases.
- [What is Apache Iceberg?](https://spice.ai/learn/apache-iceberg): Apache Iceberg is an open table format for large analytic datasets. Learn how Iceberg works, its architecture, key features like schema evolution and time travel, and how it compares to Delta Lake and Hive tables.
- [What is BM25 Full-Text Search?](https://spice.ai/learn/bm25-full-text-search): BM25 ranks full-text search results using term frequency, term rarity, and document length. Learn the formula, work through an example, and evaluate lexical search in practice.
- [What is Data Acceleration?](https://spice.ai/learn/data-acceleration): Data acceleration caches frequently accessed data locally for sub-second query performance without moving data permanently. Learn how it works, acceleration strategies, and when to use it.
- [What is Data Virtualization?](https://spice.ai/learn/data-virtualization): Data virtualization provides a unified view of data across multiple sources without physical replication. Learn how it works, how it compares to ETL, and when to use it.
- [What is Delta Lake?](https://spice.ai/learn/delta-lake): Delta Lake is an open-source storage layer that brings ACID transactions, schema enforcement, and time travel to data lakes. Learn how Delta Lake works, its architecture, how it compares to Apache Iceberg and Hudi, and how Spice connects to Delta tables for federated queries.
- [What is DuckDB?](https://spice.ai/learn/duckdb): DuckDB is an open-source, in-process analytical database designed for fast OLAP queries. Learn how DuckDB works, its columnar architecture, and how Spice uses DuckDB as a data accelerator engine.
- [What is Hybrid Search?](https://spice.ai/learn/hybrid-search): Hybrid search combines vector similarity search with keyword matching to deliver more accurate results than either method alone. Learn how hybrid search works, ranking algorithms like RRF, and when to use it.
- [What is LLM Inference?](https://spice.ai/learn/llm-inference): LLM inference is the process of generating text by running input through a trained large language model. Learn how inference works, key performance metrics, and optimization techniques like KV caching, quantization, and speculative decoding.
- [What is LLM Tool Calling?](https://spice.ai/learn/llm-tool-calling): LLM tool calling is a capability where a model outputs structured function calls instead of plain text, enabling AI agents to query databases, call APIs, and take actions. Learn how tool calling works, security considerations, and how MCP standardizes tool use.
- [What is Reciprocal Rank Fusion (RRF)?](https://spice.ai/learn/reciprocal-rank-fusion): Reciprocal Rank Fusion (RRF) is a ranking algorithm that merges result lists from multiple search methods by position rather than score. Learn how RRF works, its parameters, and why it is the standard fusion method in hybrid search.
- [What is Retrieval Augmented Generation (RAG)?](https://spice.ai/learn/retrieva