AlchemyLake Now Installs Directly Into Your Databricks Workspace
Pull-quote: “One bundle deploy, and the workspace gains a creative studio: the app for people, MCP for agents, and ai_render() for anyone who lives in SQL.”
The announcement
AlchemyLake now installs directly into a Databricks workspace. The integration ships as an Apache-2.0 Databricks Asset Bundle: one databricks bundle deploy, and your workspace gains three ways to turn governed data into sealed deliverables.
This is the integration we designed AlchemyLake around from the start. The platform’s whole premise, numbers computed by code, verified after generation, sealed to source, assumes a governed data estate underneath. Databricks is where those estates live.
The three paths in
1. The app, for people. A Streamlit app runs inside the workspace behind your existing SSO. Analysts bind Unity Catalog tables or Genie spaces and render reports, decks, infographics, and briefings without leaving Databricks. Deployment is the standard Databricks Apps model on Premium or Enterprise workspaces.
2. MCP, for agents. Thirteen governed tools register with Genie, Agent Bricks, or any MCP-capable client, including Claude and Cursor. An agent can carry a Genie conversation and hand the governed result straight to a render tool: ask the question, get the sealed board deck, one turn. Genie conversations keep their thread identity, so multi-step research stays in one governed session, and Genie spaces get health scoring and live certification along the way.
3. ai_render(), for SQL. An optional Unity Catalog Python UDF lets anyone who lives in SQL request a sealed narrative from a query result. It runs on serverless compute with external access, and it is there for the simplest possible integration story: no new UI, no new client, just a function call.
What the bundle is, and is not
The repository is deliberately thin: an asset bundle definition, the app, and the UDF. It holds no model weights, no proprietary rendering code, and no secrets. It is an auditable client over the same MCP and REST contract the public platform exposes, which is exactly what a security review wants to read. For teams with residency requirements, a no-egress text tier routes narrative inference to your own Databricks Foundation Model endpoint, so bound rows are the only thing that ever leaves, and only for the render you asked for.
Why this matters for the practice
Our Databricks modernization practice ends every engagement at the same question: the platform is governed, now what do the humans ship from it? This integration is our answer, and it is also the proof that the answer works: AlchemyLake is a production system built by the same team that runs the practice.
The bundle is on GitHub under Apache-2.0. Start at the AlchemyLake platform page, or bring it up with us in a modernization assessment.