Mosaic AI

What it does

Databricks AI takes in enterprise data from across multiple systems, along with a choice of models ranging from classical ML to GenAI, and produces AI agent systems that answer questions, automate tasks and run in production. Teams can prepare data, build agents, deploy them through Model Serving, evaluate output quality with AI judges and human feedback, and govern access, rate limits and lineage through Unity Catalog.

People use the platform to ground agents in their own data rather than relying on a single generic model. Listed examples include FactSet's text-to-code knowledge agent for nontechnical users, Block's seller-operations automation and ICE's agent answering customer questions from proprietary financial data. The distinct pieces are built-in evaluation and governance across models, including those hosted outside Databricks.

The platform suits data and AI teams at enterprises that need accurate, domain-specific applications and can work within the Databricks workspace. Access requires a Databricks account; a 14-day trial with up to $400 in free credits is offered. Costs are usage-based, measured in DBUs, and some Foundation Model Serving rates vary by model, token type and cache use. Regional processing adds a 10% uplift on labeled models.

What it can do

• Build AI agents grounded in enterprise data with Agent Bricks
• Built-in agent evaluation with AI judges and human feedback via Agent Framework and MLflow
• Deploy agents, chains and GenAI or classical ML models with Model Serving
• Govern agents end-to-end with Unity Catalog access controls, guardrails and lineage
• Create vector indexes for RAG with AI Search
• Fine-tune open source LLMs or pretrain custom models with Model Training

Works with

• Models named on the page: OpenAI GPT-4, Meta Llama 3, Anthropic Claude 3

Example workflows

Build a knowledge agent over enterprise data

Input: enterprise data across systems. Steps: prepare data and create a vector index with AI Search; build an agent with Agent Bricks or Agent Framework; evaluate outputs with AI judges. Output: a deployable agent answering domain questions, per the page's described approach.

Deploy and govern an agent

Input: a built agent and chosen models. Steps: deploy through Model Serving; apply Unity Catalog access controls, rate limits and guardrails; track lineage from data to deployment. Output: a governed endpoint serving the agent.

Before you choose

• Foundation Model Serving costs are DBU-based and vary by model, token type, cache reads and provisioned throughput.
• Pricing displayed does not guarantee regional availability; check product availability by cloud provider.
• The $400 credit and 14-day trial are promotional and not a permanent free tier.

Listed under

• Recruiting

Get started

Trial includes up to $400 in free credits; a free self-paced GenAI Training course is also available.

1. Create a Databricks account and start the 14-day trial from the product page.
2. Complete the free GenAI Training course or follow the Get Started Guide tutorial.
3. Prepare source data, then build and evaluate an agent using Agent Framework before deploying with Model Serving.

Final price equals SKU price per DBU multiplied by DBU rate per million tokens or hours. Regional processing adds a 10% uplift on labeled models. Trial credits are time-limited.

14-day free trial.

Source: databricks.com/product/pricing/foundation-model-serving, checked 28 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 28 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Databricks AI: build and deploy AI agent systems ↗www.databricks.com
• Foundation Model Serving | Databricks ↗www.databricks.com
• Databricks documentation | Databricks ↗www.databricks.com

Company website

databricks.com: product, pricing and policy pages

YC directory mirror

Batch and company name

Web search

Confirms the official site

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 28 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.