Half of the 400 stars were bought: Limelit Open, a GEO company, used an open-source library to do GEO

Limelit Open is an open-source, self-hosted "AI visibility" tracker. You bring your own API keys, it asks ChatGPT, Claude, Perplexity, Gemini and Google AI Overview, and it counts whether your brand shows up in the answers, where it ranks, and who it cites. In 56 days the repo picked up about 400 stars, and over the last 7 days it added +155, so it looks like it's gaining traction. Broken down, roughly the first 190 stars and nearly all 172 forks came from the same batch of freshly registered empty accounts, backed by a paid press release and a ring of GEO sites linking to it. It's a traffic funnel for the commercial product limelit.co, not an open-source project that developers found on their own. This slot is already crowded too.

Sep 29: 108 stars and 109 forks in one day

In 2026, on Sep 11, an account called limelit-geo that had been registered the day before created the repo (later renamed limelit-co/open). By Sep 28 it had 44 stars and 43 forks. On Sep 29, the day it published macOS and Linux binaries on GitHub, it jumped to 152 stars and 152 forks in a single day (per the star history on githubrepo.cloud).

Stars and forks rose in lockstep, which almost never happens with a normal open-source repo, where forks are about a tenth of stars. Here the ratio is 45%. Look at the accounts that forked it:

• The timing clusters: Sep 28 had 36, Sep 29 had 109, Oct 19 had 25, and they bunch up in the same hour (for example, 25 of them in the 12 noon hour on Oct 19), matching the day's star jump of +26.
• The names look mass-generated: LanaCarson, FelixAndrade, Heather1Murphy, Tina1Johnson, Joyce11Keeling, and so on.
• Of the 172 fork accounts I could still look up, 167 were registered between 2026 and 6 September, and only 1 have more than 5 followers. Of the 5 I spot-checked, all were registered on the same day in Aug 30, all have 0 followers, and most have a couple dozen other repos that are also forks.
• Another 14 accounts have already been banned or deleted by GitHub.

GitHub API, checked 11-06. This is the classic "stars plus forks sold as a package" botting service.

GitHub stars and forks.

The stretch after Oct 31 is the only part I can't explain: about twenty stars a day, very evenly spread, with forks flat. It could be real traffic from the "open-source AEO tools" listicle pages, or they may have switched to a service that only buys stars and not forks. GitHub no longer publishes the list of users who starred a repo, so I can't check individually. The steady daily rate looks more like the second option (an inference, with moderate confidence).

What it actually is

It's a single Go binary with no Docker, no Node and no Postgres. On first launch it creates its own SQLite database and serves a dashboard at localhost:1515. You enter a set of questions users might ask an AI, and it periodically asks seven AIs: ChatGPT, Claude, Perplexity and Gemini through their official APIs with web search enabled, and Google AI Overview, AI Mode and Bing Copilot, which have no API, through scraping. It then calculates your visibility and share of voice against competitors, average ranking, and whose sites get cited.

This goes by four names in marketing circles: AEO, GEO, AIO and LLMO. The README spells out all four on the first screen.

The execution is serious. Every number is labeled with its sample size. Every AI answer can be opened in full, with the brand mentions highlighted. The demo site demo.limelit.co runs the repo's code as-is, and the banner shows the current commit. It ships an MCP server with 14 read-only tools and a Claude Code plugin marketplace, so /plugin marketplace add limelit-co/open installs it. The README also includes a comparison table against commercial platforms, a glossary and an FAQ. In Oct 4 a GEO content site published a long source-code audit, which confirms the code is readable.

The last command is the reason it exists: limelit upgrade --key <cloud key> moves all your local data to the paid limelit.co cloud version. Cloud plans start at $19.99 in prepaid credits (limelit.co/pricing).

Beyond stars, what else was promoted

• Paid press release: In Oct 8 ABNewswire published "Agent-first platform Limelit releases Limelit Open…", which a string of news sites such as StreetInsider and FinancialContent republished. PRWeb also ran a piece on "automating GEO for AI agents with 211 tools." ABNewswire's homepage itself says "paid press release distribution, guaranteed pickup on 500+ high-authority news sites."
• Listicles and directory pages: cloro.dev's "2026 best AEO tools" lists it among 7 open-source tools. It's on repocloud with one-click deploy, and automated directories such as openalternative and ossradar include it. From the outside, there's no way to tell which of these were paid.
• Discord, 44-second short videos, and a demo site.
• No community buzz: Hacker News has no results, nobody has mentioned it on X in the last 60 days, and Reddit search comes up empty too.

The purpose is easy to guess. A GEO company sells the promise of getting AI to mention you in its answers. Press releases, comparison tables, listicle sites and a high-star GitHub page are exactly the kinds of pages AI search engines like to cite. They're doing GEO on themselves, and the open-source repo is the highest-weight page in that set.

How crowded this slot is

Crowded. The commercial players include Profound (publicly reported 2025 Series A and B totaling $55M), Peec AI (Series A of about €21M), Otterly, and Scrunch (from $250/month). Ahrefs and Semrush have also bundled similar features. There are at least seven or eight open-source "self-hosted trackers," and nearly all follow the same playbook: open-source core plus a cloud version.

GitHub stars, 11-06.

The last line says the most. Canonry and Limelit Open have almost the same idea and architecture. One has a few hundred stars and the other has almost none. The difference isn't the product, it's the promotion budget. The biggest player in this space is the SEO/GEO library packaged as "skills" (open-seo-mcp-skills, for example, has 8,800 stars). The "self-hosted tracker" slot is stuck at a few hundred stars for everyone.

If it were you

Verdict: D. These stars weren't discovered by developers. A company bought them and pushed them with press releases. The slot already has seven or eight open-source repos and dozens of VC-funded commercial platforms, and Canonry, with the same positioning but no paid promotion, has barely gotten any stars. Trying to build a reputation by making yet another open-source AI visibility tracker isn't worth it.

Who it suits: small teams already building GEO or SEO tools that have a paid product and want an open-source entry point to bring in customers, or people who want to learn how to write a README that works as a landing page.

Who it doesn't suit: solo developers hoping an open-source project will gain stars and reputation organically, and anyone without a promotion budget or an audience of their own.

How to do it (if you still want to build in this category): stop building a dashboard that tracks everything. (1) Take the piece it does poorly. It covers 11 data sources with only 6 of them done well, so build a small tool that only scrapes the three engines without APIs (AI Overview, AI Mode and Copilot), or do the same tracking for Chinese AIs (Doubao, Kimi, DeepSeek, Yuanbao). I haven't seen anyone own this in open source yet, though I haven't checked each one (inference). (2) Change the form: ship it as a Claude Code or Codex skill or MCP that the tools call directly, not a web dashboard. Skill-style repos nearby have more than ten times the stars of the trackers. (3) Copy its README practices: run the demo on unmodified code and label the commit, label every number with its sample size, keep the comparison table honest, and put all four synonyms on the first screen.

How to get users: in your Show HN and X launch posts, include a real side-by-side screenshot of how AI answers questions about your product versus competitors. Also publish the repo as a Claude Code plugin marketplace to use this new distribution channel. And write a data piece like "we tested a popular category and these 10 sites are what AI cites most" because that spreads further than product descriptions. Don't buy stars: stars and forks arriving in the same hour, accounts registered on the same day, and forks making up 40% of stars are easy for experienced people to spot, and the numbers drop once the accounts get banned.

How much time you have: for the Chinese AI slot, probably a few months, before GEO companies in China bring the same open-source playbook over. For a general English-language tracker, the window has already closed.

What would show you're wrong: (1) After November, Limelit Open keeps adding twenty to thirty stars a day for several weeks without new press releases, and people start discussing it on Hacker News and X unprompted, which would mean the product is pulling people in on its own. (2) Canonry or another unpromoted competitor naturally reaches a thousand stars within a few months, which would mean the slot isn't saturated. (3) Public data shows limelit.co's cloud version gained a meaningful number of paying users through the open-source repo, which would mean the "open-source core plus paid acquisition" route is worth copying for small teams.