Databricks Mosaic AI — Company Dossier
Overview
**Databricks Mosaic AI** is the AI layer of the Databricks Data Intelligence Platform, designed to help teams build, evaluate, govern, and serve production-grade AI agents on enterprise data. It unifies agent authoring, retrieval, tool use, evaluation, vector search, guardrails, and governed serving behind a single gateway—tightly integrated with Databricks’ **Unity Catalog**, **Delta Lake**, **MLflow**, and **Databricks SQL** for consistent quality, lineage, and access control.
Product hub: [Artificial Intelligence on Databricks](https://www.databricks.com/product/artificial-intelligence)Mosaic AI capabilities: [Docs overview](https://docs.databricks.com/aws/en/generative-ai/guide/mosaic-ai-gen-ai-capabilities)Why Mosaic AI for Enterprise Agents
**End-to-end governed stack:** From data to deployment, everything inherits cataloged permissions and lineage via [Unity Catalog](https://www.databricks.com/product/unity-catalog) and [MLflow](https://mlflow.org/).**Production readiness:** Evaluations, tracing, safety policies, and audited serving are built in.**Flexible models and tooling:** Use Databricks-hosted models (e.g., DBRX), open models (Llama, Mistral), or external providers (OpenAI, Azure OpenAI, Anthropic, Amazon Bedrock) through a unified interface.**Lakehouse-native RAG:** Tight integration with Delta tables, documents, and SQL warehouses accelerates grounded assistants and analytics copilots.Explore the vision: [Build compound AI systems faster with Mosaic AI](https://www.databricks.com/blog/build-compound-ai-systems-faster-databricks-mosaic-ai)
Core Capabilities
**Mosaic AI Agent Framework:** Build RAG and tool-using agents in Python; log runs with MLflow; deploy to serving.Learn more: [Agent Framework product page](https://www.databricks.com/product/machine-learning/retrieval-augmented-generation) and [Create an agent (docs)](https://docs.databricks.com/aws/en/generative-ai/agent-framework/create-agent)**Agent Evaluation:** Rubric-based and AI-judge scoring, side-by-side comparisons, and trace analysis for quality assurance.Details: [Agent Evaluation docs](https://docs.databricks.com/aws/en/generative-ai/agent-evaluation/)**Vector Search:** Managed embeddings and indexes integrated with Unity Catalog for secure, scalable retrieval.Docs: [Vector Search](https://docs.databricks.com/aws/en/generative-ai/vector-search)**Model Serving:** Serverless or provisioned endpoints for real-time and batch inference with model monitoring.Docs: [Model Serving](https://docs.databricks.com/aws/en/machine-learning/model-serving/)**AI Gateway:** Central policy, safety, rate limits, usage logging, routing, and observability across models and providers.Docs: [AI Gateway](https://docs.databricks.com/aws/en/ai-gateway/)See the end-to-end guide: [Build and deploy production-quality compound AI systems](https://www.databricks.com/blog/mosaic-ai-build-and-deploy-production-quality-compound-ai-systems)
Architecture and Governance
**Data governance:** Inherits permissions and lineage via **Unity Catalog**, ensuring consistent entitlements across agents, features, indexes, and models.**Experimentation and tracing:** **MLflow** provides run tracking, model registry, and GenAI tracing for chains/agents.**SQL-native assistants:** Build analytics copilots over **Databricks SQL** and apply **AI Functions** for LLM tasks directly in SQL.Docs: [AI Functions in SQL](https://docs.databricks.com/aws/en/large-language-models/ai-functions)Models and Providers
**Databricks-hosted models:** Including first-party **DBRX** and curated open models (Llama, Mistral).**External providers via Gateway and External Models:** OpenAI, Azure OpenAI, Anthropic, Amazon Bedrock, and more—accessed through a common interface with central policy and logging.Model catalog: [Foundation models overview](https://docs.databricks.com/aws/en/machine-learning/model-serving/foundation-model-overview)Common Use Cases
**RAG knowledge assistants** grounded in Unity Catalog data and enterprise documents**Support and operations assistants** that call tools and business systems**Analytics copilots** over Databricks SQL, Delta tables, and BI dashboards**Code assistants** for internal repositories**Document extraction and summarization** for contracts, tickets, and logs**Multi-agent flows** for enrichment, planning, and tool orchestrationGet started: [Agent Framework tutorial (notebook)](https://docs.databricks.com/aws/en/generative-ai/tutorials/agent-framework-notebook)
Integrations and Ecosystem
**Frameworks:** LangChain and LlamaIndex with MLflow tracing[LangChain on Databricks](https://docs.databricks.com/aws/en/large-language-models/langchain)[LlamaIndex with MLflow tracing](https://docs.databricks.com/aws/en/mlflow3/genai/tracing/integrations/llama_index)**Observability:** MLflow 3 GenAI tracing across apps[LangChain integration (tracing)](https://docs.databricks.com/aws/en/mlflow3/genai/tracing/integrations/langchain)**BI and SQL:** Power BI, Tableau via Databricks SQL warehouses; LLMs via SQL AI Functions[AI Functions in SQL](https://docs.databricks.com/aws/en/large-language-models/ai-functions)**Partner ecosystem:** Example—Arize for GenAI observability with Agent Framework[Arize + Mosaic AI guide](https://arize.com/blog/harnessing-databricks-mosaic-ai-agent-framework-and-arize-for-next-level-genai-applications/)Who It’s For
Data platform teams standardizing on the LakehouseML engineers shipping LLM apps with governance and evalsApp developers building RAG and tool-using agents on enterprise dataAnalytics leaders rolling out AI assistants across BI and operationsCustomer Adoption
Databricks reports 15,000+ organizations on the platform, with notable names including Block, Comcast, Condé Nast, Rivian, and Shell.
Company profile: [Databricks on LinkedIn](https://www.linkedin.com/company/databricks)Buyer Considerations (User Sentiment)
ProsStrong end-to-end environment for data engineering, ML, and GenAI; MLflow registry supports collaborative MLOps.Reviews: [G2 Databricks Platform](https://www.g2.com/products/databricks-data-intelligence-platform/reviews)Quick start for Spark/data pipelines; robust scaling once tuned.Community: [Reddit discussion](https://www.reddit.com/r/dataengineering/comments/11p2dqg/how_good_is_databricks/)Active investment in agents with built-in demos and evaluation support.Community: [Reddit thread on agents](https://www.reddit.com/r/databricks/comments/1i624w8/databricks_for_building_agents/)Lakehouse maturity reduces integration overhead for RAG and analytics assistants.Seller profile: [G2 Databricks Inc.](https://www.g2.com/sellers/databricks-inc)ConsCost awareness required; DBU spend and serving costs can add up for large jobs and always-on endpoints.Community: [Reddit discussion](https://www.reddit.com/r/dataengineering/comments/11p2dqg/how_good_is_databricks/)Learning curve for Spark, clusters, and GPUs; advanced features may not be available on free tiers.Reviews: [G2 Azure Databricks](https://www.g2.com/products/azure-databricks/reviews) and [Free Edition limits](https://learn.microsoft.com/en-us/azure/databricks/getting-started/free-edition-limitations)Some perceive marketing noise vs. real-world simplicity; occasional UI and dependency quirks in notebooks.Community: [Reddit discussion](https://www.reddit.com/r/dataengineering/comments/1e5px85/the_databricks_linkedin_propaganda/)Reviews: [Capterra Databricks](https://www.capterra.com/p/148499/Databricks/reviews/)Pricing and Free Trial
**Free options:** 14-day trial and a Free Edition to explore the platform.Start here: [Try Databricks](https://www.databricks.com/try-databricks) and [Free trial docs](https://docs.databricks.com/aws/en/getting-started/free-trial)About Free Edition: [Databricks Free Edition](https://www.databricks.com/learn/free-edition)How to Get Started
1. Review the product overview and capabilities.
[AI product hub](https://www.databricks.com/product/artificial-intelligence) and [Mosaic AI capabilities](https://docs.databricks.com/aws/en/generative-ai/guide/mosaic-ai-gen-ai-capabilities)2. Prototype an agent using the Agent Framework in Python with MLflow tracing.
[Create an agent (docs)](https://docs.databricks.com/aws/en/generative-ai/agent-framework/create-agent)3. Add retrieval with Vector Search, plug in tools, and define evaluations.
[Vector Search](https://docs.databricks.com/aws/en/generative-ai/vector-search) and [Agent Evaluation](https://docs.databricks.com/aws/en/generative-ai/agent-evaluation/)4. Choose models and deploy to Model Serving behind AI Gateway policies.
[Model Serving](https://docs.databricks.com/aws/en/machine-learning/model-serving/) and [AI Gateway](https://docs.databricks.com/aws/en/ai-gateway/)5. Observe, iterate, and compare runs with MLflow tracing and eval dashboards.
Watch a quick overview: [AI agents on Mosaic AI in 5 minutes](https://www.databricks.com/resources/demos/videos/ai-agents-on-mosaic-ai-in-5-minutes)
Additional Resources
Build an autonomous assistant: [Technical blog](https://www.databricks.com/blog/build-autonomous-ai-assistant-mosaic-ai-agent-framework)Deploy an agent to serving: [How-to guide](https://learn.microsoft.com/en-us/azure/databricks/generative-ai/agent-framework/deploy-agent)Product docs and concepts: [Mosaic AI docs](https://docs.databricks.com/aws/en/generative-ai/guide/mosaic-ai-gen-ai-capabilities)Quick Facts
Company: Databricks (HQ: San Francisco; 13,000+ employees; ~1,050,000 LinkedIn followers)Source: [Databricks on LinkedIn](https://www.linkedin.com/company/databricks)Product: Mosaic AI for agents, evaluations, vector search, model serving, and gatewayHub: [Artificial Intelligence on Databricks](https://www.databricks.com/product/artificial-intelligence)
