OpenGenie 1.0.0 — the Databricks coding agent, on the model you choose

OpenGenie 1.0.0

The Databricks coding agent, on the model you choose.

Genie Code is the best coding agent for data engineering because it can see your workspace. What it
does not let you do is choose the model, or control what the work costs. OpenGenie is an open-source
Databricks App that gives a coding agent the same kind of reach and hands those two decisions back to
you.

• Site and documentation: https://raphaelmansuy.github.io/opengenie/
• Install guide: https://raphaelmansuy.github.io/opengenie/install/
• The tour: https://raphaelmansuy.github.io/opengenie/tour/
• The evidence: https://raphaelmansuy.github.io/opengenie/evidence/

What is in it

A coding agent inside your workspace. One opencode ACP process per conversation, driven by a
FastAPI backend that serves the API and the built React front end from the same origin. The agent
reaches the workspace through a per-session MCP tool server: Unity Catalog, SQL warehouses,
notebooks, jobs, Lakebase and Databricks Apps, 28 tools in the core profile and 41 in the server
(README §"What the agent can reach").

The model is a setting. 13 models from 4 providers ship in the catalogue — Databricks serving
endpoints, OpenAI, Anthropic and Google Vertex — and the endpoints your workspace exposes are
discovered at run time. Personal provider keys are envelope-encrypted with a key held in a Databricks
secret.

Writes stay inside the fence. The tool server is started with one folder, one schema and one
job-name prefix. A write anywhere else is refused by that server with the reason; a write made from
the app outside the folder waits for your approval.

Done means proven, and work is durable. A claim is matched against the tool calls the agent
actually made. Turns run on the server and are journalled in Lakebase, so closing the tab is safe and
a restart states what was interrupted instead of replaying it.

A notebook, not a chat log. Cells with their own Ask and Run, Run all that stops at the first
failure, a run map, Fix with agent that sends the engine's own error, attachments and pasted images.

Jobs and pipelines as first-class surfaces. Runs, their states and the actions each state allows
are computed server-side; the browser has no write path to either, so the fence has one home.

This release states its own version

A deployed page now answers the first question anyone asks when something looks wrong: which build
am I looking at?

• apps/coding-agent/VERSION is the source, and it lives inside the app folder on purpose — a
  Workspace-folder deployment only receives git ls-files apps/coding-agent, so a root VERSION
  would never reach the container.
• GET /api/version answers the version, the commit the platform resolved (empty rather than
  invented when there is none) and the docs, source and releases links. /api/health answers
  {"ok": true} for every build ever deployed, which is exactly how three stale deployments were
  once reported as successes; this route cannot.
• The version prints in the top bar beside the product name, in the account menu with its commit, and
  on the home screen's This workspace panel. Each is a link to the documentation site.
• make -C apps/coding-agent verify now proves it: the live app must report this checkout's
  VERSION, the served bundle must contain it, and the bundle must link the site.

Verified, not assumed

The release commit was deployed to the dev workspace (dbc-cb61f1e1-bd8e) and read back from the
deployment record: deployment 01f1dd7b25601f42b72bb54dfbeae9e0, resolved_commit
2063af36832f4a60f6c5c5a35ea05063cddd431c on main, app RUNNING, and scripts/verify-deployed.sh
reporting 20 passed, including /api/version answering v1.0.0 and the served bundle carrying
the same version and the documentation link.

Test counts at this tag: 1303 Python, 959 front-end, 30 site pages, and the site's own
11 stated-count checks, which fail the build when a count in the prose drifts from the code.

Requirements before you deploy

A Databricks workspace with app compute, the Databricks CLI signed in, a Lakebase project, and a
Databricks secret scope for the encryption key. The install page
lists every value in app.yaml that names the author's workspace and must be changed first.

Licence

Apache-2.0. Use it, fork it, run it
inside a company; keep the attribution.