Startup OS lets AI agents operate a software business: review what happened each week, decide what to try next, draft the work, and bring you the decisions that are yours to make. You supply the direction, the judgment, and the final say. It is in private beta.
The problem it solves
A software business rarely fails for lack of code. It fails for lack of sustained attention across positioning, pricing, retention, support, architecture, and planning, most of which a founder working alone cannot cover.
AI can now do much of that work to a useful standard. What is missing is structure. An agent without durable memory, defined authority, reusable methods, and a standard for evidence produces a large volume of plausible work that does not add up to anything. Startup OS is that structure.
How it fits together
Everything Startup OS knows about your product lives in your own GitHub organization, not in our database.
| Part | What it is |
|---|---|
| Your product’s repository | Your code. Startup OS files its issues here, and proposes code changes only where you allow it. |
| The Operations Instance | A second repository, named operations, that holds the product’s context, goals, permissions, and run history. See The Operations Instance. |
| The operating loop | Scheduled GitHub Actions workflows in the Operations Instance that observe, decide, and propose. See The operating loop. |
| The Startup OS GitHub App | How Startup OS acts on GitHub, within the permissions your manifest grants. It can never merge a pull request. |
| The dashboard | Where you see every product’s state and what needs you. See The dashboard. |
| The CLI | startup-os, or sos, for onboarding and maintaining a product from your terminal. See The CLI. |
The four planes
The platform is organized as four planes. Each answers a different question an agent would otherwise have to guess at.
| Plane | Provides |
|---|---|
| Context | Durable, versioned truth about each business. |
| Capability | Reusable skills and provider-neutral tools. |
| Reasoning | Goal framing, skill selection, and self-critique. |
| Governance | Decision rights, autonomy bounds, evidence standards, and an audit trail. |
What it is not
- Not an application. It operates products; it is not one.
- Not an agent framework. It is the operating discipline an agent works within.
- Not a workflow engine. Reasoning is the default, and fixed sequences are the exception.
- Not a replacement for the founder. It removes the operating load, not the judgment.
Where to start
- Quickstart: onboard your first product, in the browser or with your coding agent.
- The operating loop: what runs each week, and what lands for you.
- Permissions and operating modes: what Startup OS may do, and how you change it.
Connect strategy to daily work
The go-to-market workflow connects your accepted strategy and quarterly bets to registered tests, dated evidence, weekly decisions and the product business overview. Existing permissions and execution owners remain in force; accepting a strategy does not automatically start tests or authorize spending.