Quickstart

Create a project, understand its files, run it, and see the result on the dashboard.

Before you start

You'll need a Harumi account and access to the platform. This guide uses the web app; if you prefer working locally, see the harumi CLI guide instead — everything below has a CLI equivalent.

From your projects home, click New project and choose whether it belongs to your personal workspace or an organization. Harumi provisions a git repository for the project, seeded from a starter template.

Look at what you got

The starter template commits four files to your new repository:

harumi.toml

main.py

dashboard.toml

README.md

Open the Code tab to browse them:

harumi.toml — the run manifest. It says what command Harumi runs and on which compute kernel:
```
[run]
command = "python main.py"
kernel = "or_python_small"
```

main.py — your entry point. The starter version writes a small result to output/output.json:

result = {
    "objective": 128_450,
    "rows": [
        {"name": "Item A", "value": 52_400},
        {"name": "Item B", "value": 38_900},
    ],
}

dashboard.toml — declares the widgets shown on the Dashboard, bound by key to output/output.json. The starter config binds a metric to objective and a table to rows — matching what main.py writes.

No notebook cells

A run is exactly the command declared in harumi.toml, executed against the code on the live branch (main). There's nothing else implicit about it.

Edit and commit

Open a file in the Code tab, switch to Edit, make a change, and click Commit changes with a commit message. This writes directly to main — there's no separate save/publish step.

For anything beyond quick edits (adding real data files, working in your own editor), see Working with the repository or use the CLI to git push from your machine.

Run it

Click Run in the toolbar. Harumi clones the repository at the current commit and executes the harumi.toml command on the configured kernel.

See the result

Back on the Dashboard, View latest output shows a live terminal with stdout/stderr while the run is in progress. Once it finishes, the widgets declared in dashboard.toml render using the JSON your run wrote — so the starter project shows a real metric and table on its very first run.

Next: change the kernel or the model

To use a larger kernel or a licensed Gurobi image, edit the kernel value in harumi.toml:

kernel = "gurobi_python_medium"

Available kernels: or_python_small, or_python_medium, or_python_large, or_python_xlarge, or a gurobi_python_* size if your organization has a Gurobi license.

To build the actual model, either write the code yourself, or open Harumi AI Chat and describe the problem — in Agent mode it can build and run a model for you. See Harumi AI Chat.