Harumi

Overview

What Harumi is, how a project is organized, and where to go next.

What is Harumi

Harumi is an AI-assisted platform for building, running, and scheduling operations research and optimization models. Every project is backed by a real git repository: you (or Harumi AI Chat) write Python code against a solver kernel (OR-Tools or Gurobi), commit it, run it, and see the result on the project's dashboard.

There are no notebooks or manual file uploads to configure — your code, its run configuration, and your data connections all live in the project's repository and settings, versioned like any other git project.

How a project is organized

Every project has six sections, available from the project sidebar:

SectionWhat it's for
DashboardThe default view. Shows the latest run's results, recent runs, and any scheduled runs.
DocumentationMarkdown documents for the project — write down business rules, assumptions, or design notes.
CodeBrowse, edit, and commit the project's git repository directly in the browser.
Data SourcesConnect external databases (PostgreSQL, MySQL, SQL Server, Oracle) to query from your code.
ChatHarumi AI Chat — ask questions, have it read/write project files, or route a request straight to the solver.
SettingsRename the project, manage Secrets (environment variables for your code), and repo details.

A run is driven by two files committed to your repository:

  • harumi.toml — declares the command Harumi executes and the compute kernel (CPU/RAM and solver image) to run it on.
  • dashboard.toml — declares the widgets (metrics, tables, charts) shown on the Dashboard, bound to the JSON your run writes to output/output.json.

Where to go next

On this page