Forkbase Product Thesis

One-Line Idea

GitHub meets App Store for agent-controlled software.

Forkbase is a marketplace and iOS container app for forkable, agent-operable apps that users can install, run, inspect, customize, and remix with the help of their own AI agents.

It is closer to "GitHub meets App Store for agent-controlled software" than a generic AI app builder.

Core Problem

People want software that adapts, but most software is locked.

If a user wants one custom feature, one bug fixed, one screen changed, or one workflow optimized, they usually cannot do it. They must wait for the company behind the app, switch tools, or build something from scratch.

At the same time, AI agents are becoming powerful enough to modify software, but they are still not reliable enough to create polished, stable, end-to-end applications from nothing every time.

The opportunity is in the middle: agents should start from existing working apps, install them, understand them, make small changes, operate them, and remix them.

The Three Broken Worlds

What Forkbase Solves

Installable, forkable, ownable, agent-operable software.

• Installable like an app store.
• Forkable like GitHub.
• Ownable like source code.
• Controllable by personal agents through app tools and user-like UI control.
• Remixable with economic credit flowing upstream.

The user should open the Forkbase app first. Inside Forkbase, they can connect whichever agent they trust, such as Hermes, OpenClaw, or a custom local agent.

I want a workout app that fits my routine.

Then their agent can search the marketplace, choose a suitable app, build it with the user's Mac and Xcode, install it into the container, configure it through UI control or tools, make small source changes, and publish a fork later.

MVP Shape

Prove one loop: find, install, run, control, customize.

MVP Demo

"Install a workout tracker and create my first push day."

1. User opens the Forkbase app.
2. User connects Hermes, OpenClaw, or another agent.
3. User asks for a workout tracker and first push day.
4. Agent browses Forkbase registry.
5. Agent chooses an app, explains permissions, and builds it on the Mac.
6. IPA is imported into Forkbase Container / LiveContainer fork.
7. Agent launches the app inside Forkbase.
8. Agent uses UI control to complete onboarding.
9. If the app exposes tools, agent switches to tools.
10. User sees the app actually filled out.

The agent is not just calling APIs. It can actually use the app.

Trust Model

Do not start with ratings.

Ratings are easy to manipulate, especially if both people and agents can create accounts or perform actions. A rating system creates a trust problem before the core product is proven.

Payments

Defer payments until the install/remix loop feels real.

Agent-Operable Apps

Agents should have a ladder of control.

1. Visual control when the agent only has pixels.
2. Accessibility control when the app exposes readable UI structure.
3. Tool control when the app exposes explicit app actions.
4. Source control when the agent needs to modify, fork, or improve the app.

The agent should not need to guess everything from pixels, but it also should not be blocked if an app does not expose a perfect tool API.

Positioning

Do not position Forkbase as another AI app generator.

AI agents make software personal, but they need good existing software to start from. Forkbase gives agents a marketplace of working apps they can install, operate, and remix, while giving creators a way to earn from reusable open-source software.