New · OBD-II in Auto Center, a Virtual DJ in Audio Player →

One app suite. Make it yours.

RobOS is a suite of desktop apps, from a point of sale and an auto shop to household finances, games, music and developer tools, all built on one knowledge graph. Use the apps as they are, customize any of them to fit how you work, or describe a new app and RobOS builds it.

  • Use ready-made app suites, free and local
  • Customize any app by describing the change
  • Build new apps from context, with proof they work

Free & open source · macOS, Linux, Windows · Other install options

Optional AI features use the agent you already have, or none at all
Claude CodeOpenAI CodexGitHub CopilotGemini CLI · AntigravityLocal models via Ollama

Three ways to use RobOS

Use the apps. Change them. Build new ones.

Every app in the suite is described in a knowledge graph (KGraph): what it is, how it works and how to test it. That one description is what lets you go from using an app, to reshaping it, to building one that did not exist.

1. Use the suites

Ready-made apps for shops, families, players and developers. They run on your machine, keep your data local and need no subscription.

See the suites →

2. Customize any app

Describe the change you want. It becomes an edit to the app's KGraph, an agent builds it, and tests prove it before you keep it.

What you can change →

3. Build a new app

Describe an app, or import what you already have. RobOS models it in the KGraph and routed agents build it, the same way the suites were built.

How apps get built →
Household · change the appExample
Add an allowance envelope for each kid that tops up every Friday and shows on the accounts page.Ctrl+↵ submit
✦ Plan the change

The change lands in the app's knowledge graph first

  • + Envelope
  • + TopUpSchedule
  • ~ AccountsPage
  • + test: Friday top-up

An agent builds it on a branch, runs the app headlessly and hands back a narrated walkthrough. You keep it or throw it away.

The app suites

Suites to start from, not stop at

Each suite was modeled in the KGraph, built by agents and verified by automated browser or gameplay tests. Use them as they are, or ask for the version that fits you.

Browse every app in the directory, including the developer suite →

Customize existing apps

Every app is a starting point

Because each app is described in the KGraph, a change is a request, not a rewrite. Describe it in plain words; RobOS turns it into an edit to the graph, an agent implements it on a branch, and tests and a narrated walkthrough show you the result before you keep it.

Point of Sale
Suggest a pastry to anyone who orders an oat-milk latte before 10.

A change to your register, built and tested against your menu.

Household
Flag any subscription that went up in price since last year.

A new check over your reconciled transactions.

Auto Center
Explain these trouble codes to the customer in plain English on the estimate.

A new section on the estimate, written from the scan on the ticket.

Tabletop RPG
Add a side quest for tonight with a talking wolf in it.

New content for a game whose rules already live in the graph.

Developer suite
Show which services a pull request touches before I review it.

A new panel for the developer apps, from the same graph.

Any app
Make the text bigger and put the buttons I use most on the left.

Small changes count too. You decide what your copy of the app looks like.

These are example requests. How well a change goes depends on the agent you use, and you review the evidence before you keep anything.

AI-powered features, always optional

Inside the apps, AI is an extra you switch on

Apart from building and changing apps, many actions inside the apps have optional AI-powered features. Each one hands your request to the agent you choose, along with the slice of the KGraph it needs, so the answer is about your data. Leave them off and the apps work the same.

Example · Audio Player

Create a Virtual DJ

Describe a host, and the agent writes a short intro for each song a few songs ahead. RobOS Voice speaks it between songs, on your desk or on a Google Cast speaker. The same player can build an AI Playlist from a description of a mood.

  • Off unless you turn it on
  • Pick the agent, the voice and how often it talks
  • Works from your own library, not a streaming catalog

Read how we built it →

Virtual DJ dialog with the AI agent set to Antigravity, the DJ described as a big loud radio host, a voice, how often the DJ talks and a DJ volume slider at 150% of the music
The Virtual DJ dialog: pick an agent, describe your DJ, choose a voice.

Optional, never required

The apps work without AI. AI features are buttons you choose to press, and the API mode for quick requests is off until you turn it on.

Your agent, your account

RobOS runs the agent CLI you already signed in to, or a local Ollama model. No new subscription.

Grounded in your graph

The agent gets the part of the KGraph it needs, so it answers about your data instead of guessing.

Guarded by default

A prompt security guard checks every request for secrets, personal data and prompt injection before it leaves the app.

Build new apps from context

Describe it once. The graph does the remembering.

Behind every RobOS app is a KGraph: a standard description of what exists, how it behaves and how to verify it. That is context engineering made durable. With the context written down, building and rebuilding is cheap work that routed, lower-cost models can carry, and every change comes back with proof.

  1. DescribeA sentence, a sketch, or an import of the repos, APIs, databases and docs you already have.
  2. ModelRobOS turns it into KGraph nodes: entities, screens, rules and the tests that define "working".
  3. BuildAgents get only the slice they need and write the code on a branch.
  4. ProveThe real app runs headlessly; you get a narrated walkthrough and test evidence to review.
One change, from idea to merged PR, in the real RobOS apps
Task created. Start from the app as it is today Task created. Describe the goal and let an AI agent plan it Task created. The plan comes with acceptance criteria and test cases Task created. Synced to the tracker as real issues Task implemented. Pick a task and let an agent implement it Task implemented. The agent branches, edits, tests and pushes Task implemented. Branch pushed. No pull request yet Owner reviews the changes. Review every changed file Owner reviews the changes. Ask the agent for a change, in plain English Owner reviews the changes. The agent edits the file, runs the tests and commits Owner reviews the changes. The diff updates on its own Owner checks the evidence. Evidence is tied to a commit Owner checks the evidence. Ask for fresh evidence, plus a new case Owner checks the evidence. Evidence appears as the agent makes it Owner checks the evidence. Fresh evidence for this commit, with the new case Owner takes the walkthrough. A guided walkthrough, one checkpoint at a time Owner takes the walkthrough. The agent pops up the running app Owner takes the walkthrough. Ask for a change while the app is open Owner takes the walkthrough. The page updates live, and the change is committed Draft PR created. Create the pull request as a draft Draft PR created. The PR exists, as a draft Owner watches CI. CI runs on the pushed commit Owner watches CI. All checks pass, so the PR can leave draft Moved to ready for review. Ask for review Moved to ready for review. Now in review Reviewer approves. The reviewer opens the same pull request Reviewer approves. The reviewer approves Owner merges. The author merges the approved PR Owner watches CI and acts. CI keeps watching after the merge Owner watches CI and acts. Back to step 1: the loop starts again Owner watches CI and acts. The feature is live
1 / 31

Recorded from the real apps. GitHub, the coding agent and CI results are scripted stand-ins; the repository, tests, diff and evidence are real.

Under the hood

One knowledge graph behind every app

The same platform that builds the suites is in your hands: the graph, the importers, the agent runner and the review tools. Developers still get the full lifecycle, from planning to merge, with semantic diffs and IDE review.

One standard KGraph format

A single schema for every app, service, contract, schema, runbook and test. Whatever it is, it has a node and a known shape.

Import anything

A KGraph parser and import skills pull in repos, APIs, databases, docs and tickets and turn them into linked entries.

Viewer, editor, e-learning

Every node gets a viewer, a form editor, generated documentation and e-learning, all customizable to your domain.

Proof, not promises

Agents run the real app headlessly and hand back a narrated walkthrough and test evidence, so you review what happened rather than guess.

Review you can understand

The graph knows what a change should touch, so review starts from intent: semantic diffs, inline comments, and IntelliJ or VS Code when you want them.

Safe by construction

Disposable sandboxes, the prompt security guard and a full audit log keep agents inside the lines.

  • GitHub
  • GitLab
  • Google Drive
  • Google Cast
  • Ollama
  • PostgreSQL
  • MongoDB
  • Elasticsearch
  • Apache Tika
  • Docker
  • Jira
  • and many more

Make money with RobOS

Start a small software business on top of it

RobOS is free and open source, which makes it a ready-made foundation for a services company. Use the existing apps as your starting inventory, customize them for each customer, and earn from the gap between what customers pay for old software and what it now costs you to replace it.

1. Set up an LLC

Form a small company that owns the customer relationships, the contracts and your extensions. RobOS stays open source, and your business is the work you do on top of it.

2. Take, extend or build apps

Start from an existing RobOS app, ask for the changes a customer needs, or build a new one from a KGraph. Routed, cheaper models do the bulk of the work, so your margin stays healthy.

3. Migrate customers over

Import the systems customers run today into the graph, so moving them onto the RobOS-based version is documented and testable.

4. Sell support contracts

Recurring support covers each customer's tickets, and the KGraph keeps every fix quick to find and verify.

Use it, change it, or build your own.