why viberary exists
A library for software built by engineers who orchestrate agents.
Something changed in how software gets made. Developers now plan work, hand it to AI coding agents, review what comes back, and steer the whole thing toward a product. That's real engineering effort, but its results are scattered across tweets, repos and demo videos. Viberary collects them in one place and records the tools behind each one.
the idea
Capture what agent-driven engineering actually ships
Every listing on Viberary is a piece of software someone shipped by directing AI coding agents: a CLI, a SaaS product, a game, a library. Alongside the screenshot and links, we record the harnesses they used and the models those harnesses ran, as pairs, because "Claude Code with Opus" and "Cursor with Sonnet" are different setups.
Over time that turns into something useful: a picture of what developers are really building with Codex versus Claude Code, which models end up in shipped products, and which projects took several agents working together.
two ways to build
App builders and harnesses are different crafts
There are broadly two ways people build software with AI today, and both are good. The difference is who drives and where the code lives.
app builders
Lovable, Bolt, v0 and similar
You describe what you want, and the platform builds and hosts it. They're fast, friendly, and have brought millions of people into building software. That's a big win.
coding harnesses
Claude Code, Codex, Cursor, OpenCode and similar
Agents work inside your own repository, terminal and editor. You break down the problem, run several agents, review diffs, run tests, and keep full ownership of a codebase you can keep developing.
Our working assumption is simple: software engineers tend to go straight to harnesses. Viberary is built around that second craft, so its work isn't lost among prompt-to-app demos, and so the people doing it can learn from each other. It isn't a ranking of better and worse. It's a choice of focus.
what belongs here
What goes on the shelf
- yesApps built with Claude Code, Codex, Cursor, OpenCode, GitHub Copilot's agents, Cline, Aider, Gemini CLI or similar harnesses.
- yesApps built with several harnesses or several models; that's often the most interesting part.
- yesApps that started in Lovable, Bolt or v0 and were then developed further in a harness.
- yesWeb, mobile, desktop, CLI, libraries and open-source projects, anything with a public URL.
- not quiteApps generated entirely inside an app builder without direct code development.
- not quiteSoftware written by hand with only occasional AI autocomplete.
We don't check whether you're a professional engineer, and we don't count AI-written lines. Every submission is reviewed by a person before it's published, mostly to keep out spam and placeholder projects.
what we record
The toolchain behind every app
Each listing has one or more coding setups: a harness, a model, and an optional version or note, for example Claude Code → Claude Opus (Opus 4.5). Builders can add as many as they used. If a tool picks the model automatically, "Unknown / auto-selected" is a valid answer.
Builders can also tell the story of their process in their own words: how they split the work with the agents, how they reviewed and tested it, and what they're proud of. It appears on the app's page as "The process", because how something was built is half the interest.
This information is self-reported. We present it as what builders say they used, not as verified usage data. New harnesses and models are added to the catalog as they appear.
faq
Questions, answered
Is Viberary against Lovable, Bolt, v0 and other app builders?
Not at all. App builders have opened software to far more people, and that's a good thing. Viberary simply focuses on a different craft (engineers orchestrating coding agents in their own codebases) so the two kinds of work aren't mixed together. If you started in an app builder and continued in a harness, your app is welcome.
Do I need to be a professional software engineer?
No. What matters is how the software was built: with AI coding agents doing meaningful work in a codebase you can keep developing directly. Students, indie hackers and open-source maintainers are all welcome.
How much of the code has to be written by AI?
There's no quota. We don't police the exact definition of vibe coding or measure human versus AI lines. The agents should have played a meaningful role; an app written by hand with occasional autocomplete isn't the focus here.
What's a harness?
The tool that runs a coding agent against your code: Claude Code, Codex, Cursor, OpenCode, GitHub Copilot's agents, Cline, Aider, Gemini CLI and others. The harness decides how the agent reads files, edits code, runs commands and talks to a model.
Is the harness and model information verified?
No. Builders report it themselves, and Viberary doesn't independently verify it. Every submission is reviewed by a person before it's published, but that review checks the app, not its toolchain. If you don't know which model ran, choose "Unknown / auto-selected".
What kinds of apps can I submit?
Anything you've shipped: web apps, mobile and desktop apps, CLIs, libraries, developer tools, games and open-source projects. It needs a public URL (a website, a download page or a public repository) and a screenshot.
Does it cost anything?
No. Listing, browsing and voting are free. There are no paid placements or sponsored rankings.