What Houston isn't.
Houston is very good at one shape of problem because it's not trying to be good at every shape. Here's the positioning, stated plainly.
What Houston is for
Houston is for building local, AI-native apps that replace the role of a junior employee for one user or one small team.
Examples of what fits:
- A bookkeeper who handles 200 clients.
- A research assistant tracking 100 sources.
- A project manager running a dozen projects.
- A content writer organizing a few dozen drafts.
- A personal CRM for a freelancer.
The unit of scale is "what one person does well." Houston aims to make one person 10x as effective, not to make a company more efficient.
What Houston isn't for
Houston is not for:
- Multi-tenant SaaS. There's no server. There's no "users" table. There's the user.
- Massive datasets. Houston uses the filesystem as the source of truth. Fast for thousands of entities. Slow for millions.
- Real-time collaboration. It's a personal tool. Two humans don't share a workspace.
- Offline-first apps. Houston assumes Claude Code is available. Without the AI, the app is a UI with nothing behind it.
- Apps where AI is optional. If your users should be able to do everything without the AI, you're building a normal app, not an AI-native one.
These aren't bugs. They're positioning choices. Houston is very good at what it's for because it's not trying to be everything.
The scale ceiling
A Houston workspace is fast up to a few thousand entities. Past that, filesystem queries start to drag.
A simple test: "can one person reasonably care about all this?"
- If yes, Houston fits. 50 clients, 100 projects, 300 contacts — all fine.
- If no, Houston is the wrong tool. 50,000 contacts across a sales org isn't an assistant problem, it's a database problem.
The ceiling matches the product. Houston is for apps at human scale, because the AI in Houston is operating at human scale.
Cost and privacy
Houston is free and open source. The user pays for the LLM calls directly — they bring their own Claude Code setup or API key.
This means a few things:
- There's no subscription fee from Houston.
- There's no server that sees your users' data.
- Cost scales with the user's own usage, under their own account.
- The AI still sees what it processes — that's between your user and Claude, not a problem Houston solves.
If your users have sensitive data and strict compliance requirements, they need to read the Claude terms of service and decide. Houston doesn't make that decision for them.
When to pick something else
Pick something else if:
- You're building multi-tenant SaaS.
- You need real-time collaboration.
- You're storing millions of records.
- Your users need to work offline.
- AI is a feature, not the interface.
Pick Houston if:
- You're building a tool that replaces or augments a specific role.
- Your user is a person, not an organization.
- Edge cases are common and impossible to predict ahead of time.
- You want your users to bend the app to their needs without calling you.
- You'd rather ship capabilities than workflows.
Scaffold a project with pnpm create houston-agent
and start with one page. Ship it. Use it yourself. Teach it
things. That's the fastest way to understand what AI-native
actually feels like in practice.