Does Craft Anything include an AI model?
The inspection and comparison tools work without AI. Building uses a coding agent installed and signed in on your computer, with Claude Code as the first fully documented path.
Getting started · Updated November 2026
Use Craft Anything by choosing a website, app or game, describing the result you want and letting a coding agent work through Study, Build, Check and Refine. Clean room is the default; Deep mode is optional and should be used only on software you own or are authorized to inspect.

Install the Craft Anything desktop app and sign in to a supported coding agent on the same computer. Claude Code is the most complete path documented by the project.
The fastest installation route is ArtCraft Launcher, which selects the correct build for macOS, Windows or Linux and keeps it updated. Standalone installers and packages are also available, including a browser build and a headless command-line tool.
Craft Anything does not replace your coding agent. It provides the project workflow, inspection tools, approval layer and checks around an agent you already control. That distinction matters: model usage and sign-in remain attached to your own Claude Code, Copilot CLI, Grok or ACP-compatible agent.
Craft Anything moves through four visible stages: Study, Build, Check and Ready. Each stage produces project artifacts you can inspect instead of a single opaque chat response.
In Study, the app records what it can observe about the target. For a website, that includes screenshots, visible text, measured layout, colors, typography and opened menus. For an app or game, it can use a window picture or deeper internal tools when the selected mode allows them.
In Build, your coding agent creates a separate project from the request and the captured evidence. Check compares the result with the original; websites are evaluated at three screen widths. Ready opens the result for review. You can then request changes or continue working in the generated folder.
Choose Clean room when you only need observable behavior. Choose Deep only when you own the software or have explicit permission to inspect its code, files or runtime internals.
Clean room is the safer default. It refuses tools that would read the original’s code or files and works from what a visitor or user can see. That includes screenshots, visible text, layout measurements and interaction states. It is a product rule, not an operating-system sandbox.
Deep mode unlocks source, file, JavaScript and binary inspection tools. It is appropriate for private maintenance, migration and learning where authorization is clear. Craft Anything does not bypass logins, DRM or anti-cheat systems, and the generated build should not copy logos, photography or text unless permission exists.
Run the agent’s normal sign-in first, then select the supported backend in Craft Anything settings. Claude Code is the verified first path; other ACP agents are documented with different maturity levels.
Craft Anything invokes the coding agent on your machine rather than receiving a copied credential. Your Claude Code, GitHub Copilot, Grok or other ACP tool keeps its own permissions and account relationship. This keeps model billing and account access outside the Craft Anything project database.
Agent behavior still matters. A coding agent can run commands and edit files with the permissions you grant it. Craft Anything asks before unfamiliar command categories and keeps approval rules visible, but the operating-system boundary remains the source of truth. Run generated projects in a folder and environment you trust.
Review the checks, open the generated project and request focused changes. You do not need to discard the project and start again for every mismatch.
Each project folder keeps the request, captures, generated app, checks, investigation evidence and open questions. That structure makes the work inspectable after the chat has moved on. If a page differs from the reference, the next request can point to the specific section, breakpoint or behavior that needs attention.
The command line exposes the same engine for repeat work. A rebuild can start from a URL or from a plain description, and an existing project can receive follow-up instructions. Teams can automate the workflow through the CLI or expose the tools to another agent through MCP.
Craft Anything is early alpha. It is ready for evaluation, learning and controlled projects, not for assuming that every inspection engine or agent adapter has equal maturity.
The desktop app, CLI and MCP server are usable today. Website rebuild checks, Claude Code building, Godot, Unity and Unreal game rebuild paths, Deep mode and Ghidra integration are documented as working. Other agent adapters and some behavior checks are still partial or tested against stand-ins.
The repository publishes current status and tool-by-tool notes. Read those before relying on an untested engine or backend. For business-critical work, keep a human review step, verify generated dependencies and scan the resulting project before deployment.
These are the practical questions people ask before choosing a workflow or starting a rebuild.
The inspection and comparison tools work without AI. Building uses a coding agent installed and signed in on your computer, with Claude Code as the first fully documented path.
It runs a guided multi-step workflow rather than one unverified generation. You still review approvals, inspect the checks and request refinements.
No. Clean room controls which inspection tools Craft Anything offers. Your coding agent and operating system retain their own permissions.
Compare the plans, create your account and continue into the Craft Anything workspace. The desktop app remains available separately for local inspection and agent work.
Product and competitor claims were checked against first-party documentation on November 16, 2026. This article compares documented workflows; it does not claim independent benchmark results.
Corrections: hello@storyteller.ai