Status
In private development. Early access by invitation.
This is a first session with Emolysis, step by step, so you can judge whether it fits how your team works. It goes from a local file to a tested gold table and a dashboard, then to a pipeline on Databricks.
Every use of Emolysis needs an Emolysis account, local files included; the Free plan is one. The session has two halves. The first half needs no platform: a file, a question and DuckDB. The second half connects Databricks and builds a pipeline. You can stop after step 6 and still have a working dashboard on your machine.
1. Sign in and connect a model
On first start, the app asks you to sign in to Emolysis first, with your GitHub, Microsoft or Google account in the browser. Then it takes you through six short steps before any conversation opens: a welcome, how the agent should reach a model, connecting it, a test, your default model and an optional sample question. To reach a model, you paste a key from the model provider you choose or point the app at a compatible endpoint of your own; on a paid plan with included model usage, you can use your Emolysis account instead. Every plan accepts your own key, and a key you add is kept in your machine's credential store. The test sends one short request and shows the reply and how long it took; it counts like any other request on that route, and the first conversation opens only after a model has answered. See Model keys and credits.
2. Create a project
Choose New project, or run emo init orders in a terminal. Emolysis creates a folder with models/, pipelines/, tests/ and an emolysis.yml. Put the folder in a Git repository if you want pull request checks later. See Projects and emolysis.yml.
Shell
emo init orders3. Start local, with no platform
Drop a CSV or Parquet file into the project, for example a year of orders. Emolysis works on local files with DuckDB alone, which is enough for this part. No platform is involved, so no warehouse compute is used. The agent's prompts still go to the model provider you choose.
4. Ask a question
Type: "Monthly revenue by region, with order counts." The agent reads the file's schema, proposes a semantic model and asks you to confirm anything it cannot infer, such as which column holds the order date.
A semantic model here is a short list: revenue and order count as measures, month and region as dimensions, and the grain of the answer. Read it before you confirm. This is where a wrong answer is easiest to prevent. See Semantic model and Grain.
5. Watch the shape and the tests
The agent picks Aggregation / gold, fills its template and runs the golden tests on DuckDB. You see the generated SQL, the test results and, if a test fails, the fix the agent makes before it runs the tests again. See Golden tests on DuckDB.
6. See the dashboard
The dashboard does not wait for the gold table. As soon as you confirm the definitions, the agent answers with a draft: a KPI for revenue, a line for monthly revenue by region and a table for the top regions, drawn from illustrative rows generated from those definitions. It then checks each chart against the tables in your project's lakehouse and links it to the Aggregation / gold template from step 5. The charts go live on the gold table once it is tested and deployed. You can change a chart type or a label. See Dashboards from a question and Chart types.
7. Connect Databricks
Choose Connect, then Databricks. The connect flow first checks your Emolysis sign-in, then you authorize with Databricks' own sign-in and pick a workspace and the compute that deploys should use. The Databricks profile loads, and your catalog appears in the app. See Connecting Databricks.
8. Build a pipeline
Ask: "Keep history of the customers table, tracking region and segment." The agent picks SCD Type 2, asks which column is the business key, fills the Databricks template and runs the golden tests on DuckDB. You can pull a small sample of the real table to see the result on your own data.
9. Deploy
When everything is green, choose Deploy. Emolysis packages the pipeline as a Databricks Asset Bundle and deploys it to the workspace you picked. If you prefer a review first, the project goes to a pull request instead, where emo check runs in your CI. See Deploying.
10. Share
Publish the dashboard and choose who can see it: anyone with the link, your organization, or only you. The dashboard leaves your machine when you publish it, and only then. See Sharing dashboards.
What to read next
- Pipeline shapes explains the ten shapes and how the agent chooses between them.
- Golden tests on DuckDB explains what the tests prove and what they do not.
- Data handling lists exactly what leaves your machine, and when.