ArtCraft has open-sourced Craft Anything, a native Rust application that tries to turn software reconstruction into an inspectable agent workflow rather than a single chat with a coding model. The storytold/craft-anything repository appeared on GitHub on October 31, 2026 and is offered under either the MIT or Apache-2.0 licence. The project labels itself early alpha.
The product loop is study, build, check and refine. A user points Craft Anything at a website, desktop or phone app, game or file set and says what they want made. The engine runs inspection tools that gather evidence about how the target is built, then hands construction to a coding agent. Claude Code is the first builder named in the documentation. The intended output is a separate clean-room project in a stack the user requests, not an in-place edit of the original.
Craft Anything says its engine exposes on the order of 320 tools across native programs, managed runtimes, mobile stacks, Electron and JavaScript, websites, firmware, captures, databases and game formats. Deep mode can call Ghidra when it is installed. The same engine powers a desktop app, a CLI and an MCP server, so other agents can drive the same inspection surface. Status and activity views, plus a build-activity log meant to hide secrets, are part of the stated workflow so a person can see which command or file is in play.
Governance is part of the design claim. Commands outside the tool's rules are supposed to surface as an approval card before they run. The repository also states that Craft Anything is for software a user owns or is permitted to examine, and that it does not bypass DRM, logins or anti-cheat. Those are declared product boundaries rather than a legal opinion or a proof of enforcement.
Version 0.3.0 landed on November 2, 2026 with packaging and installer work across major desktop platforms. New Runtime covered the release as a harness-pattern story: the interesting idea is the comparison and approval loop around the agent, not another one-shot generator. The project still calls a broader public release unfinished and does not publish fidelity benchmarks, so reliability across real targets remains unproven.
For agent-tool builders the useful takeaway is operational. Reconstruction work needs evidence gathering, a plan, observable execution, an external comparison check and a clear gate for consequential commands. Craft Anything makes that loop public to inspect while remaining early software.