Resources

Self-paced courses are free with a Databricks account. The instructor-led ones are paid.

Official demo of governing agents through Unity Catalog and Unity Gateway; the page exposes only its title, so check the content before relying on details.

Official Agent Bricks demo video; the page exposes only its title, so check the content before relying on details.

Official video on training and fine-tuning with AI Runtime; the page exposes only its title, so check the content before relying on details.

Where most features are announced, a year before the documentation settles. Recordings are free.

Walkthrough of a metastore, catalogs, schemas and grants in the UI.

The two incremental ingestion paths side by side, on the same data.

A declarative pipeline built and run end to end, expectations included.

How the task graph, triggers and dependencies fit together.

What Lakeflow actually does underneath the declarative syntax.

Why bundles exist and what they replace.

Past the getting-started example: variables, targets and real deployments.

Dataset to dashboard to Genie space, in one sitting.

The product thinking behind AI/BI, useful before you build one.

What serverless changes about sizing and cost.

How the warehouse decides to scale, and what that does to the bill.

An independent first look, with the caveats the launch post leaves out.

What Photon accelerates, what falls back, and whether it pays.

Clustering shown through the operation it helps most.

A conversation about where the layers go wrong, not a happy-path demo.

Index types, embeddings and when vector search is the wrong tool.

What changed in MLflow 3 and why runs and models were separated.

Authoring, evaluating and serving an agent, end to end.

Official channel: Data+AI Summit sessions, product deep dives, release walkthroughs.

Independent channel with hands-on Databricks walkthroughs and opinionated reviews of new features.

The official skills that teach Claude Code, Cursor, Copilot and Codex how to work on Databricks today. Installed and kept current with databricks aitools install.

Installer for the agent skills plus an MCP server with 40+ Databricks tools and a builder app with Claude Code built in. The MCP server is best-effort, not an officially supported product.

The notebooks behind the official courses. Course material goes stale, so check the last commit date before following one.

The open-source catalog and its REST API. Reading the API is the quickest way to understand what a securable really is.

Metadata-driven bronze and silver pipelines: one specification per dataset, one generic Lakeflow pipeline that reads it. Formerly dlt-meta, and one of the few Labs projects with an official docs page.

Cost estimator for a workload before you run it: DBU sizing, cloud cost and configuration comparison, deployed as a Databricks App.

In-process Spark quality rules with quarantine and statistics tables, rules kept in a table and alerting to Kafka or email. The closest non-Databricks equivalent of DQX.

Python bindings for Deequ, "unit tests for data" on Spark: constraint suggestion, a metrics repository and anomaly detection over the metric history.

Checks written in YAML against warehouses and Spark, now positioned as a data contracts engine.

Expectation suites plus generated data documentation across Spark, pandas and SQL backends. Stewardship moved from the Great Expectations organisation to Fivetran in May 2026.

Schema and statistical typing for pandas, Polars and PySpark: declare a DataFrame schema and validate it in the job and in CI.

Time-series API on top of Spark: as-of joins, lagged values, resampling and rolling statistics without hand-written window functions.

Runs one operation across many Unity Catalog tables at once. Quiet since 2025 and part of its classification API is deprecated: useful as a script, not as a dependency.

Lightweight SQL execution over the Databricks SDK, for tools that need to run a query without a Spark session.

No longer actively maintained: its own README points to Databricks Asset Bundles. Worth knowing because plenty of repositories still deploy with it.

Toolkit for migrating a workspace onto Unity Catalog: assessment, table migration, rewriting jobs and dashboards.

Data quality checks on PySpark DataFrames and tables, batch and streaming, with quarantine and summary metrics. The programmatic counterpart to pipeline expectations.

Automates migration onto Databricks from other warehouses: source profiling, SQL dialect conversion and row-level reconciliation. Formerly called Remorph.

Generates synthetic data with Spark. The fastest way to get a realistic table to practise on.

Pytest fixtures for Databricks: real workspace objects created and cleaned up around a test.

The shared scaffolding behind the Labs projects: config, logging, CLI wiring for a Python project on Databricks.

The dbt adapter. Relevant if your transformations live in dbt rather than in pipelines.

Terraform for workspaces, clusters, policies and grants. Bundles ship code, Terraform provisions the platform.

Read and write Delta tables from Rust and Python without Spark. Useful for understanding the protocol.

The VS Code extension: sync a local folder, run a file on a cluster, debug with Databricks Connect.

The SQL connector for Python: query a warehouse from anything that is not a notebook.

Index of the Labs projects. Exploratory tools, not covered by a support agreement.

One-command demo notebooks for most Databricks features. The fastest way to see a concept running.

The source behind the demos: worth reading even if you never install them, because it is idiomatic current-generation code.

Reference implementations by industry. Useful for seeing a full pipeline rather than a snippet.

The CLI itself, including the bundle implementation. Read the issues when a bundle behaves oddly.

Working bundle definitions for jobs, pipelines and apps. Faster than assembling one from the reference.

Everything the REST API can do, typed. What to reach for once shell scripting the CLI stops scaling.

Delta Lake itself. PROTOCOL.md is the authoritative description of the transaction log.

Upstream Spark. Worth opening when a Databricks doc says 'see the Spark documentation'.

A real workspace you can keep, at no cost. Every concept on this site can be tried in it; a few enterprise features cannot.

One command installs a working notebook, pipeline and dashboard for a topic into your own workspace. The fastest way to see a feature run rather than read about it.

Unofficial, MIT-licensed set of 71 Databricks product icons taken from official artwork, in SVG and PNG with Mermaid and draw.io packs. Handy for architecture diagrams; read the Databricks brand guidelines before public or commercial use.

Quality checks on DataFrames, from Databricks Labs. Unsupported, so read it as a pattern to copy as much as a dependency to take.

The book on the table format itself: transaction log, time travel, clustering, streaming reads. Free from Databricks in exchange for an email; also on O'Reilly.

Free ebook, registration required. Patterns rather than syntax.

Learning Spark, second edition. Free from Databricks with registration.

Delta Lake: The Definitive Guide. Free with registration.

Free online notes on Delta Lake internals. Deeper than the docs, occasionally behind them.

Free online notes on Spark SQL internals: Catalyst, physical plans, adaptive execution.

Shorter and more hands-on than the Definitive Guide. Good first book if you have never used Delta outside Databricks.

Vendor-neutral: what a lakehouse is made of and which decisions are irreversible. Read it before designing a platform, not while fixing one.

Recipes rather than theory, by a Databricks solutions architect. Useful when you know what you want and not how it is spelled.

Declarative pipelines end to end, from ingestion to serving. Written before the Lakeflow rename, so it says Delta Live Tables throughout.

The standard introduction to the DataFrame API and Structured Streaming, by four people who built them. Spark 3.x, so the APIs are current.

No Databricks in it at all, and that is the point: the vocabulary and the trade-offs that every platform shares.

Chip Huyen on building with foundation models: evaluation, retrieval, fine-tuning, and what breaks in production. The closest thing to a textbook for the GenAI exam's harder half.

Feature stores, training pipelines, serving, drift. Vendor-neutral, and the reason the ML exam's system-design questions stop being guesswork.

A worked case of tuning ai_extract field by field against reviewed ground truth, showing why a fix for one field can raise errors overall.

The research behind agent memory: why it keeps raw session traces apart from distilled memory, and how a background pipeline keeps the store consistent, with the benchmark method.

Shows how the managed storage location is inherited from metastore to catalog to schema, how SET MANAGED LOCATION changes where new tables land, and where an external table's data goes when converted to managed.

Where a feature is explained before the documentation covers it. Marketing-heavy; the engineering posts are the ones worth the time.

Other people's errors, with answers. Unverified by definition: check anything you find here against the documentation before relying on it.

Curated old-name to new-name catalogue. Also feeds the release-notes job.