routerhubSI/uncensored-ai-api · routerhubSI

13 days, 600 stars, not a single line of code: an "uncensored AI API" repo that turns out to be an ad posted on GitHub

What it is A cheap, uncensored LLM API reseller using GitHub as an ad page. There is no code in it.

Why it took off It didn't really take off. Stars and forks came in batches from throwaway accounts within half an hour, plus a ring of copycat repos and impersonated commits. The goal was to make the GitHub page rank for the "uncensored AI API" search term and get cited by AI search.

DWhat you can take from it Don't copy the star-buying or the impersonation. If you have a real product, take only the clean half: exact search terms in the repo name and opening lines, FAQs written as short answers plus an llms.txt, repos split by long-tail keywords, and utm-tagged links.

Grade D: the stars were bought, and bought stars don't earn reputation or followers. Star-buying and impersonated commits also risk getting the repo taken down. The only usable lesson is the SEO playbook, and 7API and UncensoredLabs already run the same template. Almost nobody knows about it yet

Data (updated daily) · Repo created 2026-10-24 · 428 stars · 7 days +60: GitHub · stars
Data 2026-10-26 → 2026-11-08 (14 days, 14 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

routerhubSI/uncensored-ai-api looks like an open-source project that just took off: created 13 days ago, 602 stars, 53 forks. Look closer and there's no code, only a README, a dozen-plus docs, a seo/ directory and an upload manual. It's a search landing page for the API reseller routerhub.si, and most of the stars and forks were bought. If you want your open-source project to earn stars, there's little to learn here about building reputation. What you can learn is how a GitHub repo gets used as a search entry point, and how to spot fake stars.

Nov 2 noon, 23 minutes

2026 Nov 2 12:06 to 12:29 (UTC), this repo received 3 commits with the author name "Peter Steinberger" and his email, that of well-known developer steipete. GitHub ties commits to accounts by email, so the commits were attributed to steipete's own account, and his avatar appeared in the repo's contributors list. All 11 commits carried this attribution, and 8 of them were backdated to Oct 5 to 24, earlier than the repo's creation date (10-24 16:55).

In the same period, 12:14 to 12:30, 11 forks came in. The fork accounts look like yusufverhoeven3328, MagdalenaLacroix987, alainfransson683, LuciaSiekiera808: a foreign first name, a surname, then a string of digits, the usual format for bulk-registered throwaway accounts. Stars went from 368 to 503 that day, and a sister repo under the same organization, glm-5.3-abliterated, went from 395 to 541 the same day.

How the stars came

Date	stars	fork	What happened
10-24 → 10-26	0 → 341	6	Two days after creation, 341 stars, and almost no links pointing to it could be found online.
10-26 → 11-01	341 → 368	6	A week of no movement, with a drop back to 366 in between.
11-02	503(+135)	30	Those 23 minutes above
11-06	602	53	Another wave early the next morning, 11-05

(GitHub daily star counts; fork times from GitHub's fork list)

Forks arrived in batches: of the 58 forks, 24 crowded into the 1 minutes from Nov 09:15 09:50–35, 11 landed in the 2 minutes on Nov 15, and 11 came on Nov 5 05:48–06:18. The stargazer list page and API for the repo couldn't be accessed, so we couldn't check individual stargazers. The claim that the stars were bought rests on three things: the fork accounts, the time clustering, and the synchronized growth across several repos. It's a strong inference.

On the other side, real discussion is almost nonexistent: searching "routerhub" on X for nearly 30 days turns up nothing; searching "routerhub reddit" only returns its own repo, its X account, and a few AI tool directory sites; its X account @RouterHubSI has posted 16 times in total.

What it actually sells

routerhub.si is a prepaid, per-token model gateway compatible with the OpenAI and Anthropic APIs, with "40 live models" on its homepage. It sells two things: low prices, and "uncensored." The site lists Claude Opus 5.5 at $0.72 input and $3.59 output per million tokens, and the page itself says "82.0% less"; Grok 4.7 is marked "96.4% less". It also offers uncensored and abliterated (refusal-removed) versions of DeepSeek, Qwen and GLM (routerhub.si homepage, 11-06). It accepts cryptocurrency and has no subscription.

Scamadviser rated it on Oct 6 as "very young", and Gridinsoft gave it a trust score of 50/100.

Their audience is clear: the repo's topics include claude-code, cursor and opencode. Coding agents burn through a lot of tokens in a day, and developers wanting cheap Anthropic-compatible keys have been around for a while.

How the repo is used as an ad

The opening of seo/serp-and-ai-search-plan.md states the goal:

"rank for uncensored AI, uncensored AI API and uncensored AI model on Google and DuckDuckGo, and be cited by AI answer engines (Google AI Overviews, ChatGPT, Perplexity, Copilot, Gemini)."

It lists a search-term map ("uncensored AI API pricing" maps to docs/08, "uncensored AI API key" maps to docs/04) plus ten page rules, such as "the primary keyword must appear in the first 100 words", "give a short answer first, since AI engines cite the first two sentences", "each FAQ answer 40–60 words", and "external links only to RouterHub, Telegram and sister repos". The repo also includes an llms.txt for AI crawlers. The upload manual says "Do not initialize with a README (the kit already has one)", which suggests a complete ready-made template you just upload. The first screen of the README has three buttons: visit the site, join Telegram, get an API key, all with utm_source=github links.

Around it sits a ring of other repos:

Long-tail repos under the same org: openrouter-alternative (riding OpenRouter's brand name), uncensored-vs-abliterated-ai, uncensored-models, glm-5.3-abliterated.
Single-repo alt accounts: Andreas698/glm-5.3-uncensored (107 stars), PierreTucker147/deepseek-uncensored (65 stars), martina9516/uncensored-ai-models (103 stars). Each account has only this one repo, and all their profiles point to routerhub.si.
AI tool directories: EveryDev.ai and SkillsLLM both list it.

It works, to a degree: searching "RouterHub uncensored API" returns almost all of its own repos in the top ten (Brave search). GitHub has high domain authority and fast indexing, and AI search often treats it as a trusted source, so a few hundred stars can put an "authoritative-looking open-source project" on the results page for a commercial keyword.

Who else uses the same template

Repo	Repo created	Scale	Compared with alternatives
7API/uncensored-ai-api(7api.st)	10-22	245 stars / 63 forks, with stars slowly declining	Nearly word-for-word the same description, and there's a glm-5.3-abliterated repo that came out two days earlier
UncensoredLabs/uncensored-ai-api	11-01	5 days, 840 stars, one day peaked at 558	Same name, same playbook, homepage shows "6.4B tokens processed"
wideeyededi/uncensored-nsfw-ai-video-api	—	—	Same tactics, switched to NSFW video
PurpleDoubleD/locally-uncensored	—	3,391 stars	A genuinely working local uncensored setup, using abliterated weights from Hugging Face + Ollama

These three are very likely the same team, or they bought the same SEO outsourcing template (inference, not confirmed). The stars on 7API fell from 223 to 213, which matches what fake-star cleanup on GitHub looks like.

As a product, "one key for all models" is OpenRouter's shape, and "uncensored" is a positioning Venice.ai has been using for a long time. It has nothing new of its own.

If it were you

The take: Don't copy it. These 600 stars aren't reputation, they're rent on an ad slot. You can buy stars and build one by mass-fake-starring and submitting under other people's identities, but no one will get to know you because of it, and GitHub may ban your account and delete the repos.

Who it's for: Anyone who already has a real product (SaaS, API, tool site) and wants an extra search entry point can borrow its "repo as landing page" approach. It's also useful for people who want to understand GitHub's trending list without being fooled by fake stars.

Who it's not for: Developers hoping to collect stars and build a résumé through open source. What you need is real people starring, real people sharing, and a project you can put on a CV; this playbook offers none of that. And the step of submitting under someone else's identity will directly ruin your reputation.

How to do it (only the clean half):

Use the exact term you want to be found for as the repo name, the description and the first heading, and make clear "what it is" within the first 100 words.
Write FAQ answers as short 40–60 word responses, and add an llms.txt so AI search can quote whole passages.
Split repos by "category term / specific long-tail term / competitor-alternative term", for example one covering a general question, one about a specific model, and one named "xxx-alternative".
Add utm parameters to your homepage links so you can measure how many people actually come from GitHub.
Don't touch fake stars, bulk fork from alt accounts, or impersonate someone by changing the commit author. Those three are off limits.

How it gets users: Its acquisition runs entirely on search, not community. If you copy this, you're also waiting on search traffic, not stars. For stars, you still need something genuinely useful plus one proper launch.

When to get in: No need to rush. Once there are many of these repos, GitHub cleaning up fake stars and search engines demoting GitHub pages are only a matter of time (inference). The window isn't yours to control.

What would mean you're wrong: If in a month or two the stars on routerhubSI and UncensoredLabs are still climbing, the repos haven't been touched, and the Telegram group has lots of real users talking, then "repo as ad slot" is more durable than it looks and a clean version is worth building seriously. Conversely, if the stars start falling like 7API's, or the repos get banned, the playbook has run its course.

GitHub repo ↗

Research notes: competitor table, acquisition evidence, full research report

Competitors

Name	Where	Launched	Scale	How it differs from this one
OpenRouter	Web	2023	Top model aggregation gateway (exact figures unverified)	OpenRouter: official pricing, doesn't market uncensored; RouterHub sells "uncensored + 80–90% cheaper" and opened a repo to ride the openrouter alternative brand term
Venice.ai	Web	2024	Known privacy/uncensored AI service with an API (unverified)	Venice self-hosts open-source models with a public brand; RouterHub also resells closed-source big-lab models
NanoGPT / Featherless / ArliAI / Infermatic	Web	2023-2025	Small/mid pay-as-you-go API (unverified)	Products are similar; RouterHub's only difference is its GitHub SEO and star-faking acquisition
Hugging Face abliterated weights + Ollama, local	Open source	From ¥2024	Free; a similar local project, locally-uncensored, has 3391 stars	Local and free but needs a GPU; RouterHub sells a hosted version
7API(7api.st)	Web	Repo 2026-10-22	Repo 245 stars / 63 forks	Nearly identical description and repo structure, accepts crypto, key doubles as login; likely the same team or the same SEO outsourcing template
UncensoredLabs	Web	Repo 2026-11-01	Repo 5 days old, 840 stars	Same repo name and playbook, also has a web chat

Where users come from

Confirmed GitHub repo used as an SEO/GEO landing page: the repo contains seo/serp-and-ai-search-plan.md, whose goal is to rank for "uncensored AI API" and be cited by ChatGPT/Perplexity/AI Overviews; it ships llms.txt; the homepage carries utm_source=github; UPLOAD_CHECKLIST.md is the upload manual
Confirmed Star/fork inflation: 24 forks arrived at 11-01 09:15-09:50, and 11 arrived at 11-02 12:14-12:30 when commits were pushed; fork accounts are "first name + last name + digits" types like yusufverhoeven3328; multiple repos in the same org gained stars on the same day; stars on the template repos slowly declined; stargazers endpoint returns 404
Confirmed Alt-account satellite repos: Andreas698, PierreTucker147 and martina9516 are all single-repo accounts, with homepages at routerhub.si?utm_campaign=…, and they published a release titled "live RouterHub CTAs"
Confirmed Identity impersonation (submitting under a well-known developer's name): 11 commits list the author as Peter Steinberger <[email protected]>, and the API attributes them to the steipete account; 8 commits are backdated to before the repo was created
Confirmed Competitor brand-term hijacking: the routerhubSI/openrouter-alternative repo
Confirmed Directory listings: EveryDev.ai and SkillsLLM have listing pages
Confirmed Telegram / X: the README has a Telegram badge; the X account @RouterHubSI has only 16 posts

routerhubSI/uncensored-ai-api research report

Date: 2026-11-06. Data sources: our github database (repo_profile / repo_day), the GitHub API (repos, commits, fork lists), the routerhub.si homepage text (Jina Reader), Brave / Exa search, and the x-radar corpus.

Bottom line

This is not an open-source project that took off on its merits. It is the SEO landing-page repo for the gray-market API reseller routerhub.si. Its 600 stars were mostly manufactured: batches of stars and forks arrived at the same moments, and the fork accounts are batches of "first name + last name + digits" alt accounts. On top of that, a set of alt-account satellite repos cross-link each other, and the commit author was disguised as Peter Steinberger (steipete). What it's after isn't the fame from stars. It wants GitHub's high-authority pages to rank routerhub.si for the "uncensored AI API" search term, and to be cited by AI search. The money comes from reselling Claude / GPT / Grok tokens at a discount (listed prices 82%–96% below official), plus "uncensored / abliterated" models.

What it is

The repo has no code (language is empty). Its content is a README, a dozen-plus documents under docs/, a seo/ directory (keyword-map, metadata, serp-and-ai-search-plan), llms.txt, and an UPLOAD_CHECKLIST.md upload manual that says "homepage carries utm_source=github&utm_campaign=uncensored-ai-api", "set 20 topics", and "pin to the org page".
The product being pushed is routerhub.si: a prepaid, per-token gateway compatible with the OpenAI / Anthropic interfaces, with "40 live models" on the homepage. Sample pricing: Claude Opus 5.5 input $0.72 and output $3.59 per million tokens, which the page itself labels "82.0% less"; Grok 4.7 is labeled "96.4% less"; there are also uncensored and abliterated versions of DeepSeek / Qwen / GLM (routerhub.si homepage, 2026-11-06).
The domain is very new: Scamadviser rated it 2026-10-06 as "very young, might be a scam", and Gridinsoft gave it a trust score of 50/100 with a note about "recent domain registration" (Brave search results).
Prices 80–90% below official usually imply gray-market sources such as pooled accounts, stolen keys, or resold trial credits (industry-common inference, unverified).

How the stars came (data card + live GitHub check)

Date	stars	forks	Notes
10-24 created → 10-26	0 → 341	6	Two days after the repo was created it hit 341 stars, when almost no external links existed yet
10-26 → 11-01	341 → 368	6	Barely moved for a week, still slipping slightly (366)
11-02	503(+135)	30	On the day, 3 commits landed between 12:06–12:29; the sister glm-5.3-abliterated repo in the same org went 395→541 (+146) that day
11-05 → 11-06	542 → 602	42 → 53	Another wave

Evidence of inflation:

Forks arrived in bulk: of 58 forks, 24 landed within 35 minutes of each other on 2026-11-01 09:15–09:50, and 11 landed on 11-02 12:14–12:30 (just 15 minutes after the push commit). 11 more came in on 11-05 05:48, and 06:18,7 on 10-24 18:03–18:31 (GitHub API /forks).
Fork account usernames: yusufverhoeven3328, MagdalenaLacroix987, alainfransson683, LuciaSiekiera808, abigailgunduz66775, kaitobrouwer529... all in the bulk-registered format of 'international first name + surname + string of digits'.
Several repos in the same org gained stars together on the same day, which doesn't look like organic spread. The stars of the sibling template repo 7API/uncensored-ai-api have been slowly declining from 223 to 213, the typical pattern of GitHub purging fake stars (our repo_day).
Both the web page and the API for the stargazers list return 404, which is rare, so we can't check the starring accounts one by one.
Almost no real discussion online: x-radar returned 0 results for 'routerhub' over the last 30 days; Brave search for 'routerhub reddit' only surfaces its own GitHub, X account and directory sites; the X account @RouterHubSI has only 16 posts.

Growth tactics (breakdown)

GitHub as SEO backlinks and landing page (main lever, confirmed): the repo name, description and H1 all use the target search term 'uncensored-ai-api' directly, with 20 topics. The seo/serp-and-ai-search-plan.md file states the goal: 'rank for uncensored AI, uncensored AI API… on Google and DuckDuckGo, and be cited by AI answer engines (Google AI Overviews, ChatGPT, Perplexity…)', and sets rules like 'First 100 words contain the primary phrase' and 'External links only to RouterHub, Telegram'. The repo also ships llms.txt for AI crawlers to read. Result: a Brave search for 'RouterHub uncensored API' returns almost only its own repos in the top ten.
Star and fork inflation (confirmed, evidence above): high star counts lift a GitHub page's search ranking and can also fool 'trending on GitHub'-style recall (our recall was fooled this way).
Sock puppet matrix (confirmed): Andreas698/glm-5.3-uncensored (107 stars), PierreTucker147/deepseek-uncensored (65 stars), martina9516/uncensored-ai-models (103 stars), each account owning only this one repo, all with homepage routerhub.si/?utm_source=github&utm_campaign=…, and all publishing releases ('live RouterHub CTAs'). The official org also opened three long-tail repos: openrouter-alternative (riding OpenRouter's brand term), uncensored-vs-abliterated-ai, and uncensored-models.
Name borrowing (confirmed): all 11 commits in the repo are authored as 'Peter Steinberger <[email protected]>', so GitHub attaches them to the steipete account (the API returns author.login = steipete). The contributors list on the repo page shows this well-known developer's avatar. 8 of the commits are backdated, from 10-05 to 10-24, earlier than the repo creation time of 10-24 16:55. It also forked litellm and litellm-docs (purpose unverified: possibly planning a provider PR to LiteLLM, or just borrowing the association).
Directory listings (confirmed): EveryDev.ai and SkillsLLM both have listing pages for it.
Telegram / X (present, but small volume): the README has a Telegram badge; X @RouterHubSI has only 16 posts.

Competitors

Name	Where	Launched	Scale	How it differs from this one
OpenRouter	Web	2023	A large head aggregator gateway for models (exact scale unverified)	OpenRouter uses legitimate channels and official pricing and doesn't push uncensored models; RouterHub sells 'uncensored + 80–90% of the price', and even opened a repo to ride the 'openrouter alternative' term.
Venice.ai	Web / App	2024	A well-known AI service focused on privacy and uncensored models, with an API (scale unverified)	Venice hosts open-source models itself with a public brand; RouterHub also resells closed models from big vendors.
NanoGPT, Featherless, ArliAI, Infermatic and others	Web	2023–2025	Small to mid-sized pay-as-you-go APIs, some with uncensored / RP models (unverified)	Same category; RouterHub's difference lies mainly in acquisition tactics (GitHub SEO + star inflation), not in the product.
Abliterated weights on Hugging Face (huihui-ai, mlabonne, etc.) + running locally with Ollama	Open source	From ¥2024	Free; PurpleDoubleD/locally-uncensored in our database has 3391 stars	Free to run locally but needs a GPU; RouterHub sells the hosted version
7API(7api.st)/ 7API/uncensored-ai-api	Web + GitHub	Repo 2026-10-22	245 stars / 63 forks	The description is nearly word-for-word identical ('one key for Claude, GPT, Gemini, Grok, DeepSeek, GLM and Kimi…'), and it also has a glm-5.3-abliterated repo; accepts cryptocurrency, and the API key is the login credential. Two days earlier than RouterHub, so it's very likely the same team or the same SEO contractor using a shared template.
UncensoredLabs / uncensoredlabs.dev	Web + GitHub	Repo 2026-11-01	5 days, 840 stars, 11-04→11-05, single-day +558	Same repo naming, same playbook, homepage advertises '6.4B tokens processed'
Various 'relay stations'	Web	long-standing	Many	All resell Claude / GPT at low prices; RouterHub adds an 'uncensored' positioning on top, targeting the English market.

Why it 'took off'

It never really took off. The 600 stars are the result of paid acquisition and self-dealing, not evidence that developers use it; forks rising from 6 to 53 also come from batch sock accounts (the recall rationale 'rising forks mean someone is integrating' does not hold).
What's actually worth studying is its GitHub→SEO/GEO acquisition model: GitHub has high domain authority, Google indexes it quickly, and AI search treats it as a trusted source. So spending a few dozen dollars on a few hundred stars can put an authoritative-looking 'open-source project' on the results page for commercial search terms like 'uncensored AI API'. The money comes back from prepaid top-ups on routerhub.si.
Timing: in 2026, coding agents like Claude Code / Codex / OpenCode are widespread and burn through tokens, so developers have strong demand for cheap Anthropic-compatible keys; its topics include claude-code, cursor and opencode.

Differentiation

At the product level there's basically no differentiation: 'one key for multiple models + OpenAI/Anthropic compatibility' is OpenRouter's standard form, 'uncensored' is a positioning Venice and others already own, and 'cheap' relies on gray-market supply. The only real difference is distribution: turning GitHub repos into SEO/GEO assets, backed by star inflation, sock puppet matrices and borrowed-name commits. This playbook isn't unique either: 7API came out two days earlier and UncensoredLabs a week later, using the same template.

Takeaways

'GitHub repo = landing page with high-authority backlinks': exact search terms in the repo name, description and H1, plus llms.txt, and FAQs written as short 40–60 word answers that AI answer engines can easily excerpt. Spread across 'category-term repo + single-model long-tail repo + competitor-alternative repo' (uncensored-ai-api / glm-5.3-abliterated / openrouter-alternative). Homepage URLs carry utm parameters, so GitHub-driven conversions can be measured.
The flip side is a lesson for our recall system too: a batch of forks landing in the same time window, fork accounts named "name+surname+number", repos with a seo/ directory but no code, homepages carrying utm tags, commit authors using celebrity emails, a stargazers list 404. All of these can go into the suspect rules. Right now our library gives it an empty suspect.

Uncertain

The starring accounts can't be verified one by one (stargazers returns 404), so the star-inflation conclusion rests on fork accounts, timing distribution and synchronized growth. It's a strong inference.
Whether routerhub.si has real paying users and what its supply source actually is remain unverified.
Whether 7API, UncensoredLabs and RouterHub are run by the same operator: only the highly consistent template has been observed, not confirmed.
Whether impersonating steipete's identity will be dealt with by GitHub (account bans or repo deletion) is unknown.

This article was written by AI from public data and the data we collect daily. Numbers are sourced, and inferences are labeled.