Open Source
Open Source

19,207 stars, still alpha: can this Rust engine rebuild a site?

storytold/craft-anything describes itself as an open-source clean-room engine in pure Rust: see how any website, app or game works, then have AI build your own version. The workflow points at a target, states what is wanted, studies the original with real inspection tools, has a coding agent build a new project, compares results against the original and iterates. A November 7, 2026 GitHub API check showed 19,207 stars, Rust as the language and a repository created on October 31, 2026. The README badges say early alpha and 100% Rust, and the license text is dual-licensed under MIT or Apache-2.0 at your option. ATTRIBUTION.md records bundled font and icon licenses while brand assets must not be modified. macOS, Windows, Linux, FreeBSD and Web are supported through a desktop app, CLI and MCP server, with Anthropic, OpenAI, OpenRouter, xAI and local models plus installed GitHub Copilot, ACP grok and Ghidra. The README admits some capabilities are built but not yet run live.

Published November 7, 20269 min read
<!-- craft-anything-clean-room-opensource-resource | open-source | 19,207 stars, still alpha: can this Rust engine rebuild a site? -->

A repository created on October 31, 2026, openly labeled early alpha, had gathered 19,207 stars by November 7, 2026. That is the question most people ask when they land on storytold/craft-anything. A direct GitHub API check on November 7, 2026 puts it at 19,207 stars, with Rust as the primary language, created on October 31, 2026 and last pushed on November 7, 2026. The README states its purpose in a single line: See how any website, app or game works, then have AI build your own version; an open-source, clean-room engine in pure Rust. The star count is loud, the status badge is quiet, and that gap is exactly why this project deserves a close look on its own.

It helps to set out how this batch divides the work, so nothing gets crossed. This article is a deep dive into one repository, storytold/craft-anything, focused on its workflow, its dual license, the boundaries of early alpha, and safety notes such as the fact that installing dependencies runs code. The five-tool AI metadata comparison in the same batch also lists it in a parameter table, but only with objective metadata and no usage walkthrough. The GA hotspot breakdown and the video API setup SOP cover the Gemini Omni video API release event and its five-step integration, a different thread entirely from this Rust repository. It also differs in angle from the DeepSeek Super-agent open-source resource dated November 2, 2026 and from aws-infinity-agent-runtime-opensource-resource dated October 26, 2026: those are a resource roundup and a runtime survey respectively, while this piece offers no checklist and no runtime comparison, only one repository's license, workflow and maturity explained in full.

What Is Actually in the Repository

In the GitHub API data pulled on November 7, 2026, the license field for storytold/craft-anything reads Apache-2.0, created is 2026-10-31T14:50:00Z, pushed is 2026-11-07T10:14:27Z, and the language field says Rust. The README badges give a second set of labels: License: MIT OR Apache-2.0, Status: early alpha, 100% Rust, and the platform string macOS, Windows, Linux, FreeBSD, Web. Both full license texts, LICENSE-MIT and LICENSE-APACHE, are in the tree alongside NOTICE, CONTRIBUTORS.md, SECURITY.md and AGENTS.md.

The egui desktop app, craft-anything-cli and an MCP server all drive one engine, parts of which compile to WebAssembly and run in a browser without the file leaving it. The README ships in nine languages, including README.zh.md.

Study First, Build Second: How the Workflow Runs

The README compresses the process into one sentence: Point it at a target, say what you want, and it studies the original with real inspection tools, has a coding agent build a new project, checks the result against the original and refines it. Unpacked, that is six steps: point at a target, state what you want, study the original with real inspection tools, have a coding agent build a new project, compare the result against the original, and refine. The project page shows this as Study, Build, Check and Refine, asking through question and approval cards when a decision belongs to you.

Clean-room is the default and the most misread term. It is a rule about what the tool offers and refuses: no reading the original's internal code, files or page source, only what a visitor can see, screenshots, visible text, measured layout, colors, fonts, spacing and opened menus. Tools that look inside are refused with a note on Deep mode, which uses every built-in engine and Ghidra when installed. Switch in Settings, run settings_set deep_mode true, or pass --deep or --clean-room for one CLI run.

One point must be explicit: clean-room is a rule about the tool, not an operating-system sandbox. It does not isolate your computer, a coding agent still acts as you, and its reach depends on its own permissions. As for what the workflow means legally, the README draws no conclusion and neither does this article. This only describes the documented workflow and is not legal advice; consult a professional for real use.

Clone It: The Commands as Written

If you want to try it, the plain clone address is https://github.com/storytold/craft-anything. The README asks for Rust 1.97 or newer, because one dependency needs it, so rustup update first; on Debian or Ubuntu the desktop app also needs a handful of system libraries. Then come three commands, reproduced as written:

sh
cargo run --release -p craft-anything
cargo install --path apps/craft-anything-cli
craft-anything-cli doctor

The first starts the desktop app, the second installs the command-line tool as craft-anything-cli, and the third, doctor, checks the install, the signed-in coding agents and the engines, reporting anything missing. Building needs a coding agent already signed in on your machine. You can also launch a rebuild straight from the command line, with questions asked in the terminal:

sh
craft-anything-cli rebuild https://example.com "Recreate this website"
craft-anything-cli commands
craft-anything-cli mcp

Every tool is both a CLI subcommand and an MCP tool, and results are evidence records with confidence, coverage and limitations; craft-anything-cli setup connects coding-agent apps.

Early Alpha: Capable, but Not a Production Tool

The repository labels itself early alpha, and the README Status block splits capabilities into three tiers. Working parts include the desktop app in Simple and Advanced modes, the CLI, the MCP server, the Study through Refine workflow with Claude Code as builder, website checks at three screen widths, game rebuilds in Unity, Unreal or Godot, Deep mode with Ghidra, and opt-in generated pictures and 3D models, relayed here only as of the November 7, 2026 check.

The partial tier matters more. The README says, verbatim, Partly there: building with API keys and other coding agents (built, not yet run live), alongside behaviour checks for apps and games and a few engine tools tested only against stand-ins. So do not look at 19,207 stars and assume the tool is ready out of the box for production work. A project labeled early alpha, openly saying parts are built but not yet run live, is best read, tested and filed issues against, not dropped into a delivery pipeline. The installers for Windows, macOS, Linux and FreeBSD on the Releases page are still marked alpha as well.

Dual License: MIT or Apache-2.0, at Your Option

The README's License and credits section says: Craft Anything is dual-licensed under MIT or Apache-2.0, at your option. The Chinese README.zh.md says the same as a dual license under MIT or Apache-2.0, your choice. That is an OR relationship, one of the two at the user's option, not a demand to comply with both licenses at once. Both full texts are in the repository: LICENSE-MIT is the short MIT license, with copyright held by 2026 ArtCraft Team and the Craft Anything contributors, while LICENSE-APACHE is the complete Apache License 2.0 body with its patent grant and NOTICE clauses. Required notices live in NOTICE.

So why does the GitHub API license field read Apache-2.0? It is the single value the platform detected, coexisting with the README badge MIT OR Apache-2.0; it does not overturn the at-your-option wording. Two asset classes sit outside this software license. Bundled fonts, icons, images and other assets keep their own open licenses, each listed with author, source and license in ATTRIBUTION.md. The ArtCraft name, wordmark and logos under docs/brand/ are trademarks governed by docs/brand/LICENSE-brand.txt, usable only unmodified and only as part of this repository; forks must remove them.

Who Builds, and What Inspects

Building needs a signed-in coding agent. The README lists Claude Code first: Craft Anything runs your own unmodified claude program with your own sign-in, never handling your account, and your use falls under your agreement with Anthropic. GitHub Copilot connects through an installed Copilot CLI over ACP, and an installed grok program through grok agent stdio over ACP. Other ACP agents can be switched per backend, and of those only Maki has been run live; a Codex adapter exists but is unverified on a real install. For questions and analysis the providers are Anthropic, OpenAI, OpenRouter, xAI, which uses Grok when XAI_API_KEY is set, and local OpenAI-compatible servers.

The inspection side backs the pure-Rust engine claim: its own decoders, disassembler and decompiler work with no outside program, covering native programs, .NET, Java, Android, Apple, Electron/JavaScript, websites, firmware, packet captures, databases and game files, with Ghidra added in Deep mode. Results are evidence records, not a guess. The egui desktop app, the CLI and the MCP server share this engine; some of it runs as WebAssembly, FreeBSD has its own release package, and the Web build is a viewer that keeps the file inside the browser.

Installing Dependencies Runs Code

The safety boundary bears on whether you can run this comfortably on your own machine. The README is blunt: Installing dependencies runs code. When a builder installs dependencies, those commands execute on your computer, and packages can run code while installing. npm runs each package's install scripts by default, and Rust build scripts and Python source packages run their own setup code. Craft Anything asks for approval before any command type it has not seen in this project. Choosing "always allow" for npm lets every later npm command run unasked, so reserve it for dependencies you trust. To install npm packages without their install scripts, use npm install --ignore-scripts.

Use it only on software you own or are allowed to examine. It never bypasses DRM, logins or anti-cheat, and the agent is told not to copy logos, photos or text without permission. For a harder isolation layer, see the Claude Code sandbox and runtime isolation SOP. How confidence-tagged evidence records are stored and reused is the other face of the codebase memory and MCP resource roundup.

Who Should Try It Now

It fits anyone who wants to read a well-structured Rust reverse-engineering engine, study how a clean-room workflow is implemented, or who already has Claude Code and Rust 1.97 and will test and file issues. The README says Simple and Advanced drive the same engine and that pure inspection needs no AI, so using it as an inspection tool with no model connected is a low-risk path.

It is a poor fit as a finished product. A repository created on October 31, 2026, still early alpha on November 7, 2026, with parts built but not yet run live, does not belong in production dependencies, and it should not promise an exact one-shot copy of any website or game. It delivers a checkable, iterable, evidence-backed rebuild process, not a finished clone. As stated earlier, this is not legal advice: reading the workflow and license texts closely beats reading the star count.

FAQ

Q1: What does storytold/craft-anything do?

A1: It is an open-source clean-room engine in pure Rust, positioned as seeing how any website, app or game works and then having AI build your own version. The workflow: point at a target, state the request, study the original with real inspection tools, have a coding agent build a new project, compare against the original, and refine. The egui desktop app, the CLI and the MCP server share one engine.

Q2: Do 19,207 stars mean it is production-ready?

A2: No. That figure is from a GitHub API check on November 7, 2026, the repository was created on October 31, 2026, and the README labels it early alpha. Building with API keys and other coding agents is partial, with the wording built, not yet run live, meaning implemented but not exercised in a real environment; behaviour checks and a few engine tools share that tier. Stars measure attention, not maturity.

Q3: Is it MIT or Apache-2.0, and must I follow both?

A3: The wording is dual-licensed under MIT or Apache-2.0, at your option, and README.zh.md says your choice: an OR relationship, one of the two at your option, not both at once. LICENSE-MIT and LICENSE-APACHE are both present in full. The GitHub API license field reads Apache-2.0 and coexists with the README badge MIT OR Apache-2.0. Bundled fonts, icons and images keep their own licenses in ATTRIBUTION.md, and brand assets follow docs/brand/LICENSE-brand.txt.

Q4: Does clean-room isolate my computer or guarantee legal safety?

A4: Neither. The README says clean-room is a rule about what the tool offers and refuses, not an operating-system sandbox: it does not read the original's internal code or files, works only from visible information, does not isolate your computer, and a coding agent still acts as you. The README draws no legal conclusion; this article describes the documented workflow and is not legal advice. To read inside, use Deep mode or --deep, or --clean-room for clean-room.

Q5: How do I avoid unknown scripts during dependency install?

A5: The README warns that installing dependencies runs code: npm runs each package's install scripts by default, and Rust build scripts and Python setup run their own code. The tool asks before any command type it has not seen, so do not pick "always allow" for a project whose dependencies you distrust. To skip npm install scripts, use npm install --ignore-scripts. Run craft-anything-cli doctor first to check the install, coding agents and engines.

Related reading

This article is AI-assisted and human-edited. Last updated: 2026-11-07

FAQ

What does storytold/craft-anything do?
It is an open-source clean-room engine in pure Rust, positioned as seeing how any website, app or game works and then having AI build your own version. The workflow: point at a target, state the request, study the original with real inspection tools, have a coding agent build a new project, compare against the original, and refine. The egui desktop app, the CLI and the MCP server share one engine.
Do 19,207 stars mean it is production-ready?
No. That figure is from a GitHub API check on November 7, 2026, the repository was created on October 31, 2026, and the README labels it early alpha. Building with API keys and other coding agents is partial, with the wording built, not yet run live, meaning implemented but not exercised in a real environment; behaviour checks and a few engine tools share that tier. Stars measure attention, not maturity.
Is it MIT or Apache-2.0, and must I follow both?
The wording is dual-licensed under MIT or Apache-2.0, at your option, and README.zh.md says your choice: an OR relationship, one of the two at your option, not both at once. LICENSE-MIT and LICENSE-APACHE are both present in full. The GitHub API license field reads Apache-2.0 and coexists with the README badge MIT OR Apache-2.0. Bundled fonts, icons and images keep their own licenses in ATTRIBUTION.md, and brand assets follow docs/brand/LICENSE-brand.txt.
Does clean-room isolate my computer or guarantee legal safety?
Neither. The README says clean-room is a rule about what the tool offers and refuses, not an operating-system sandbox: it does not read the original's internal code or files, works only from visible information, does not isolate your computer, and a coding agent still acts as you. The README draws no legal conclusion; this article describes the documented workflow and is not legal advice. To read inside, use Deep mode or --deep, or --clean-room for clean-room.
How do I avoid unknown scripts during dependency install?
The README warns that installing dependencies runs code: npm runs each package's install scripts by default, and Rust build scripts and Python setup run their own code. The tool asks before any command type it has not seen, so do not pick "always allow" for a project whose dependencies you distrust. To skip npm install scripts, use npm install --ignore-scripts. Run craft-anything-cli doctor first to check the install, coding agents and engines.

Related

Open Source

3,003 Stars, No Stable Release Yet: What Is Blocking It?

oxc-project/tsrs is a Rust rewrite of the TypeScript 7 compiler and language server, described in its README as a drop-in replacement. A GitHub API snapshot on November 8, 2026 showed 3,003 stars and Apache-2.0, but still no stable release. The README says the team aims to ship a stable version against TypeScript 7.1 with 100% conformance; aim is a target, not a delivered result. docs/STATUS.md lists GitHub prereleases, a manually assigned npm beta tag and release candidates, but nowhere declares stable. Speedup figures come from three unrelated setups that cannot be merged into one table: the project's own bench/RESULTS.md covers 19 projects on an Apple M3 Max, measured 2026-10-28 with Hyperfine medians against TypeScript 7.1.0-beta, giving 1.71x-6.62x faster; Cloudflare's up to 8x compares against TypeScript 7.0; third-party numbers near 2.7x and 5.3x have their own machines and baselines. This is a deep dive into one repository, not a multi-project ranking.

Nov 9, 20269 min read
Open Source

Announced GA, Yet Only 71 Stars: What Is Kinference?

Buoyant, the company behind Linkerd, open-sourced Kinference as an inference control plane for Kubernetes and vLLM and announced GA on November 4, 2026; a GitHub API snapshot on November 9, 2026 shows Kinference-AI/kinference (migrated from buoyantio/kinference via 301) at only 71 stars, 6 forks and 57 open issues, in Rust under Apache-2.0, created October 5, last pushed November 8, version v0.1.0-alpha.7. The README confesses verbatim that Kinference is alpha software with no sustained production exposure, lacking Anthropic-style /v1/messages, rate limiting, semantic classifiers and multi-replica support. Mechanically it exposes a gRPC interface that takes OpenAI-format requests, returns the chosen endpoint plus a mutated body, monitors vLLM KV-cache events and engine metrics, reuses NVIDIA's dynamo-kv-router crate and pairs with the Buoyant Gateway. Its vendor-reported median TTFT improvement, 6.88s to 0.31s on shared-prefix traffic, was measured against random routing rather than llm-d or Dynamo, and has no third-party reproduction.

Nov 10, 20269 min read
Open Source

AWS Infinity: an Agent Runtime Where Waiting Costs Nothing

AWS open-sourced Infinity, a suspend-and-resume agent runtime: the code lives in the hydro-project organization (aws/ and aws/hydro both 404), Apache-2.0, written in Rust, from the AWS x UC Berkeley Hydro collaboration. The official benchmark runs 80,000 agents for 20 rounds each — 3.2 million completions — at roughly 100KB of memory per agent on a single-threaded EPYC 9R14, extrapolating to about 75,000 agents in 8GB (an extrapolation; the Raspberry Pi 5 is only a memory analogy). This deep dive covers the serverless four-piece design, the simulated-model qualifier, the community critique list (durable I/O, race transfer, 25 stars and untested by the market), and how to get started.

Oct 26, 202611 min read