A "free Excel" that earned 7,000 stars in a month: GridCraft's stars didn't come from spreadsheet users, they came from the "AI clones Microsoft" news hook

GridCraft is an open-source spreadsheet written in Rust that looks almost exactly like Excel, with most of the code written by Claude agents. Created in Oct 7, it had 7,197 stars and 3,348 forks by Nov 6. But it didn't take off on its own: it's the Excel entry in ArtCraft's "AI solo-rewrites all of Adobe and Office" series, and most of its stars came from the news week of Oct 8 to 14. This piece looks at where the stars came from, what's behind all those forks, and which parts of the playbook you can learn from if you want a project that gets stars, and which parts you can't.

Two days after creation, covered by The Register

On Oct 8, 2026, UK tech outlet The Register ran a piece titled "Rust upstart reveals 'reimplementation' of Microsoft Office – with Claude on coding duty." Reporter Richard Speed opened WordCraft, the Word replacement, on an Arm Surface and concluded it "runs pretty well," but don't use it for real work. The article also consulted a lawyer: a party doing this kind of "reimplementation" can claim it never looked at the other side's source code, so there's no infringement; still, "a deep-pocketed, well-lawyered offended party" could well sue.

GridCraft had only been created that day. It was covered in the same piece as WordCraft and DeckCraft (the PowerPoint version). The next day Thurrott ran a feature on all three apps and linked the GridCraft v0.3.0 download directly.

ArtCraft, founded by Brandon Thomas ("echelon" on GitHub), is behind all of this. Before that, the same GitHub org had already shipped PhotoCraft (the Photoshop version), FilmCraft (Premiere), and a string of other apps. On Oct 4, "ArtCraft Apps – open-source Adobe compatible suite written in Rust" got 180 points and 246 comments on Hacker News. On Oct 8, Futurism ran the headline "Adobe is in trouble: someone reverse-engineered and rebuilt the whole Creative Suite."

Over half of its stars in one week

(Figures from GitHub daily snapshots, as of 11-06.)

At the peak, the org's WordCraft, DeckCraft and the artcraft main repo also topped out around Oct 13; PhotoCraft alone gained 7,184 in one day on Oct 11. It's the same crowd clicking star on each one in turn. By Nov 6, +398 over the last 7 days and +572 over the first 7 days, and still falling.

GridCraft sits mid-pack in the family: PhotoCraft 77k, artcraft 39k, PdfCraft 19.7k, FilmCraft 17.6k, WordCraft 17.4k, GridCraft 7,200. Interest in "free Photoshop" is clearly bigger than in "free Excel."

How good it actually is

The first image in the README is a quarterly sales dashboard: tables with totals rows, conditional formatting, sparklines, bar and pie charts, and it looks like Excel. It runs natively on macOS, Windows, Linux and BSD, works in the browser, uses XLSX as its own file format, and every button maps to a command that AI agents can drive via MCP or the command line.

The project grades itself in its ROADMAP (updated 11-02), and it's candid:

• Ribbon commands done: 281 of 296 (94.9%); worksheet functions: 513 of 523 (98.1%);
• But only about 50% is ready for real work;
• Still roughly 160–240 Opus 5.5 agent-hours away from beta.

ZDNet's Jack Wallen tested it on Linux, macOS and Windows on Oct 21. He said opening an Excel file with pivot tables showed "everything was in place," and startup was fast. The downsides: too few templates, and no saving as a .xltx template. His advice was to copy your file first and "do not go into the app thinking it is production-ready." On Hacker News, someone called it "very broken software produced by AI."

It shipped 16 releases in a month (v0.1.0 to v0.16.x), with about 1.92M cumulative attachment downloads. Each release has 20 to 30 platform installers, plus a launcher that pulls them automatically. That number isn't a user count, only evidence that people are actually installing it.

Where the 3,348 forks came from

For a typical popular repo, forks run around a tenth of stars; GridCraft's ratio is 46%. That's not botting. It's deliberate design:

• CineD reports that Thomas funds these free apps with income from selling AI generation credits, and pays external contributors bug bounties and stipends (exact amounts not found);
• The repo ships AGENTS.md and CLAUDE.md, a full rulebook for AI agents: claim an issue first, always write tests, don't say in PRs that an agent wrote the code;
• The result: nearly 2,000 PRs, 893 merged; over 2,800 commits across 30 days. The top contributor is Thomas himself (736 commits), followed by dozens of external contributors, many running their own agents.

People fork to submit PRs, collect bounties, and get on the contributors list. With PRs constantly landing and releases going out every day, the repo keeps looking alive, and that in turn becomes news material.

Where the money is

GridCraft is free, dual-licensed Apache 2.0 / MIT. ArtCraft sells AI image and video generation credits, with subscriptions at $10, $35 and $60 per month. Every Craft app's README opens with the ArtCraft logo, a Discord link and getartcraft.com. The free apps bring people in; the paid AI studio next to them is what earns.

But this audience cares about stars, not money: GridCraft itself doesn't earn anything. What it gives ArtCraft is reputation and traffic.

Open-source Excel alternatives have never been scarce

The ONLYOFFICE row is the one to watch: it looks like Office, has been around longer and is more stable, yet has far fewer stars than a one-month-old GridCraft. People aren't starring it because open-source spreadsheets finally got good. They're starring it because "AI copied Excel nine-tenths in a month" is itself the story.

Why this one, and why then

1. Topic. October 2026 was the first large-scale demo of "one person and their agents writing a whole desktop software suite." Media headlines were all "AI clones Adobe / Office." GridCraft is the Excel slot in that story.
2. Sentiment. Adobe settled with the US Department of Justice this March over subscription cancellation issues for $150M, and plenty of people dislike subscriptions. Microsoft Office subscriptions draw the same complaints.
3. Family traffic. PhotoCraft went viral first, and GridCraft got written into the same article the day after it was created, so it inherited attention from birth. Followers can't get that.
4. Looks genuinely usable. Screenshots look almost identical to Excel, there are installers and a web version, and ZDNet could actually try it. It's not just a concept.
5. Bounty-plus-agent-friendly repo. PRs and releases keep flowing, and fork and issue counts look unusually busy.

Plenty of people are already copying the playbook

"[Name]Craft + clean-room rewrite + pure Rust" has become a template. Since October, GitHub has added: BIMCraft (a Revit version, 949 stars), SolidCraft (a Fusion 360 version, one Chinese X post with 1,090 likes), ElyOffice (an Office suite written in Rust, 1.2k stars), Weave-Office, at least three MailCraft Outlook clones, NoteCraft (OneNote), ProjectCraft (MS Project), MatCraft (MATLAB) from the self-described ArtCraft family, and ports like wordcraft-android. Few have reached a thousand stars.

What to learn from it

• Pick a professional software product that everyone pays for and everyone hates subscribing to. The "AI clean-room rewrite" story is news in itself. But Office and Adobe are already taken.
• Launch in batches, not one at a time. Each new repo inherits attention from the family's earlier ones, and outlets will write up several in a single article.
• Write the repo as a manual for AI agents, then attach a bounty. Outside developers' agents become your labor force, and their PRs generate buzz in turn.
• Publish your own progress numbers. "Ribbon 94.9%, functions 98.1%, 160–240 agent-hours from beta." Reporters like to quote numbers like these, and readers like to share them.

If it were you

Verdict: Don't build yet another Excel, Office, or Adobe clone. Those slots are taken by ArtCraft, and the copycats have already crowded in. What's worth learning is the approach: pick a specialist application nobody has rewritten yet, whose users resent its subscription fees, launch in batches, and build the repo so an agent can pick up where the last one left off.

Who it suits: developers already active in a specialist software community (engineering, design, research, audio), who know what people in that community pay per year for which tool and what they complain about; and who can run an agent continuously for weeks and make their own demo videos and publish progress numbers.

Who it doesn't suit: anyone hoping to make money from this (this kind of project earns stars and notoriety; ArtCraft makes money from the AI subscription next door); anyone with no distribution channel who expects one repo to get noticed by the media on its own (much of GridCraft's stars came from the PhotoCraft and ArtCraft brands); and anyone unwilling to take on legal risk. The lawyer The Register interviewed was blunt: the other side has "deep pockets and lots of lawyers."

How to do it: First search GitHub to confirm the software you've chosen doesn't already have a "[Name]Craft" with over a thousand stars. Narrow the scope to one core use case and get it to look the same as the original in screenshots, with an installer and a web version. The README should state clearly that it's an independent implementation built from public specs and original assets. Add AGENTS.md and a progress table, and open some "good first issue" tickets.

How to get users: Start with Hacker News (Show HN) and the subreddit for the software you're cloning, with a headline like "I used AI to rewrite X as an open-source alternative in one month." Send screenshots and progress numbers to authors who review Linux and open-source software (the ZDNet Jack Wallen route). Spanish-language and Chinese tech accounts on X also repost topics like "AI clones [software]." Launching two or three in the same series works better.

When: the sooner the better, probably within a few months. "One person using AI to rewrite big-company software" is still news today. Once these projects are common enough that the media stops covering them individually, only quality will matter, and you'll be competing on real craft with older projects like LibreOffice and ONLYOFFICE (inference).

Signs you've misjudged it: two weeks after launch, Hacker News hasn't put it on the front page and no tech outlet has written about it; or you discover someone else has already made a "[Name]Craft" for the same software; or ArtCraft is formally sued by Microsoft or Adobe, which would make the whole clean-room playbook suddenly very risky.