The Short Answer
Craft Anything is an open-source, pure-Rust engine from the ArtCraft team that studies a website, app or game, has a coding agent build a new version, and checks it against the original. It is early alpha with 25,459 GitHub stars ten days in. When interfaces can be rebuilt, own what cannot be: on ibl.ai you own all the code and the data.
Craft Anything describes itself in one line: "understand how any software works, and craft your own." Its repository under the storytold GitHub organisation was created on 31 October 2026.
Read from the GitHub API on 10 November 2026, it carries 25,459 stars and 5,161 forks, with six tagged releases from v0.1.0 on 31 October to v0.6.0 on 9 November. It is licensed MIT or Apache-2.0.
What is Craft Anything and what does it actually do?
Craft Anything runs a four-step loop its README names Study, Build, Check and Ready. It inspects the original, a coding agent builds a new project in the language or engine you ask for, the build is compared with the original, and it is refined until it is close.
Websites are checked at three screen widths, games against their own self-test, and apps against their runs. Game rebuilds can target Unity, Unreal or Godot, among other stacks.
Under the hood is a reverse-engineering engine the README puts at 320 tools, 300 in the engine and the rest in the AI and setup layers.
Every tool is also a CLI subcommand and an MCP tool, so any agent that speaks the Model Context Protocol can drive the same engine the desktop app uses.
Building needs a coding agent on your own machine. Claude Code is supported first, running your own unmodified claude program under your own sign-in. Adapters for Codex, GitHub Copilot, Grok Build and other ACP agents exist at varying levels of testing.
Did the ArtCraft team open-source Adobe's and Microsoft's software?
No. The claim circulating on X, that "the team that open-sourced Adobe and Microsoft's stacks" built this, is wrong.
Nothing from Adobe or Microsoft was released. The ArtCraft team wrote clean-room applications from scratch in Rust in categories those companies are best known for.
The repositories say so in their own descriptions: "an open-source photo editor written from scratch in pure Rust", "an open-source, clean-room word processor in pure Rust". Most of the suite was created between 30 September and 9 October 2026.
| Repository | What it is, in its own description | GitHub stars, 10 Nov 2026 |
|---|---|---|
| PhotoCraft | Photo editor written from scratch in pure Rust | 78,382 |
| PdfCraft | PDF reader and editor written from scratch in pure Rust | 20,246 |
| FilmCraft | Nonlinear video editor written from scratch in pure Rust | 17,964 |
| WordCraft | Clean-room word processor in pure Rust | 17,729 |
| GridCraft | Clean-room spreadsheet in pure Rust | 7,387 |
| DeckCraft | Clean-room presentation app in pure Rust | 4,714 |
The distinction matters to a buyer.
A clean-room reimplementation is new code that reproduces what a product does. It is not the original product, it does not carry the original's file-format guarantees or support, and calling it "open-sourced Adobe" overstates both what was built and what was given away.
The second claim in the same post, that the tool "reverse-engineers any app from design intent", also overstates it. Craft Anything works from evidence about an existing target, and its default mode reads nothing inside that target at all.
What is the difference between Craft Anything's clean-room mode and deep mode?
Clean room is the default. The README says nothing inside the original is read in that mode, not its code, files or page source. The engine works from screenshots, visible text, measured layout, colours, fonts and spacing, as a visitor would see them. Tools that would look inside are refused.
Deep mode uses every tool, "for your private use": a program's code and files are read with the built-in engines (and Ghidra when it is installed), and a website's JavaScript is read.
The project is explicit about limits.
Clean room "is not an operating-system sandbox", and a coding agent you run still acts as you with its own permissions. Its responsible-use note says to use it only "on software you own or are allowed to examine", and that it never bypasses DRM, logins or anti-cheat.
Rebuilds are told not to copy logos, photos or text unless you confirm you own them or have permission. None of that settles the legal questions a rebuild can raise; it states what the tool will and will not do.
Is Craft Anything ready for an enterprise to use?
Not as a production dependency. The project labels itself early alpha, and its status section lists what works and what is partial.
Working today, per the README: the desktop app, the CLI and the MCP server; the full loop with Claude Code as the builder; website checks at three widths; and game rebuilds in Unity, Unreal or Godot.
Partly there: building with API keys and other coding agents ("built, not yet run live"), behaviour checks for apps and games, and some engine tools "tested only against stand-ins." Six releases shipped in nine days, which is momentum and also churn.
The practical use for most organisations in November 2026 is as a signal, not a tool. It shows how far the cost of reproducing a visible interface has fallen, and that should change how software is evaluated and priced.
How does Craft Anything relate to REA, Reverse Engineer Anything?
They are separate projects by separate authors. REA is an MIT-licensed MCP server and CLI that hands a coding agent reverse-engineering tools. It was created on 14 April 2026, passed 298,000 GitHub stars by 17 October, and carries 448,573 on 10 November 2026.
REA answers "how does this software work?" Craft Anything adds the second half of the loop: build your own version, then check it against the original. One is inspection; the other is inspection plus reproduction with a measured similarity score.
Read on the same day, the two repositories hold more than 470,000 GitHub stars between them. Developer attention to agent-driven reverse engineering is not a niche.
If an agent can rebuild an application, what is still worth paying for?
Not the interface. The pages, menus and layout a visitor can see are exactly what Craft Anything's clean-room mode is built to reproduce. A licence priced mainly on that surface, and multiplied per seat, is the wrong shape for software whose surface is getting cheaper to copy.
What clean-room mode cannot reproduce, by its own rules, is what is not visible: your records, your permissions, your audit history, and the relationships between them. A rebuilt student portal has no students in it. A rebuilt CRM has no customers.
That is where value now concentrates, and it is why the structure of your data matters more than the screens on top.
An ontology over your own systems, meaning typed relationships such as "this contact is this learner" or "this course is a prerequisite for that one", is the part a clean-room rebuild cannot see, and the records themselves never ship inside the software.
The honest caveat: none of this makes established software free to replace. Integrations, compliance, support and years of edge cases do not appear in a screenshot either, and an alpha tool that rebuilds a snake game is a long way from a production ERP.
What should a CIO do about agent-built software?
Sort renewals by what you are actually paying for. For each major licence, ask whether the price is for the interface, for the vendor's operations, or for holding your own data. The first is getting cheaper to reproduce. The third should not be rented at all.
Own the layer that cannot be copied. On ibl.ai you own all the code and the data. The platform runs inside your perimeter, reaches your systems in place through connectors and an ontology, and is model-agnostic, so the asset that matters stays yours whichever model or interface sits on top.
Treat rebuild tools as you would any agent with your permissions. Craft Anything's own README warns that installing a rebuilt project's dependencies runs code on your machine. Any team experimenting with it should do so in a sandbox, on software it owns, with the same review it gives any coding agent.
Read the primary source before the post about it. The version of this story that spread on X described something that did not happen. The repository, its licence and its status section took five minutes to read and told a more useful story.