An LLM API relay/aggregation gateway: one domain and key to call Claude, Claude Code, GPT, Codex, Gemini and more, with usage monitoring and rate limiting; formerly named PackyCode.
In June 2025 Claude Code blew up among Chinese developers, but connecting directly and paying were both hard from China. PackyCode launched then as a "Claude Code relay", and built trust by posting a public uptime monitor in the Linux.do community and answering questions in a QQ group. The real amplifier was the open-source tool cc-switch, a one-click provider switcher for Claude Code/Codex that later grew to 148k stars. In 2025-09 PackyCode was added as a preset provider in cc-switch, and on 2025-11-10 it became an official sponsor, with an exclusive rebate link (aff=cc-switch) and a 10%-off code shown in the README and inside the app. PackyAPI later turned this playbook into a "developer ecosystem partner program", using cash and credits to buy README recommendation slots from open-source projects and communities, combined with aff affiliate links, to keep catching the upstream traffic of AI coding tools.One-line growth story, compiled by us from public sources; speculation is marked separately below.
The main engine is buying placement in open-source tools plus affiliate rebates: it sponsors cc-switch (148K stars), makes itself a preset provider inside the software, and uses rebate links with an aff parameter to get people to write promotions; direct visits are 68%, with referrals and affiliates second and third.
The turbo boost is Claude Code blowing up among Chinese developers: since 2025 direct access and payment have both been hard in China, and it launched as a Claude Code relay that caught this wave of demand.
By estimate, nearly half of visits are returning users coming back to top up and check usage; about 13 of every 100 visits click through to the Stripe checkout page.
Model: Lenny's racing growth framework. Engine = paying for distribution (sponsoring the preset and README sponsor slot of the open-source tool cc-switch, an open-source sponsorship program, and aff affiliate rebates); turbo boost = Claude Code blowing up among Chinese developers; lubricant = pay-per-use billing, so existing users keep coming back to top up.
Real data: 160,900 visits, -39.56% month over month, largest channel direct at 68.45%, top three channels ranked (direct, referrals, affiliates), organic search 100% of search, only 5 search terms (packyapi, deepseek, packyapi.ai, workbuddy connect grok), top referrers (gitcode.com 5.52%, hvoyai.com 1.63%, oken.ai 1.51%), social sources (DeviantArt, X, YouTube), countries (China 58.73%, US 13.6%, Singapore 8.39%, Taiwan 7.44%, Hong Kong 6.15%), bounce rate 29.91%, 7.4 pages per visit, average stay 7 minutes 17 seconds, all from the Similarweb public page for last month; the shares of the second and third channels are not given on the public page. The 21,490 visits (about 21,500) to the Stripe checkout page come from quietmoney checkout page data.
Supporting evidence: In cc-switch's git history, in 2025 packycode was first listed as a preset provider, and later that year it was added as an official sponsor (README sponsor slots in Chinese and English, in-software promotional text, an aff=cc-switch rebate link, and a 10%-off code on the first top-up); cc-switch now has 148K stars and the sponsor slot is still there; a GitHub Gist has a long promotional article whose signup links all carry aff; the official site's /partners page has the developer ecosystem partnership program; service status is posted on the Linux.do monitoring site, and issues go to the QQ group.
Inference: The two channel circles are real; how each is split is an inference: direct typing 110,100 = returning users 75,000 + clicking links inside the cc-switch software 25,000 + promotional articles 5,100 + community recommendations 5,000; other channels 50,800 = affiliate links in promotional articles 22,000 + cc-switch's README and gitcode mirror 18,000 + community recommendations 7,800 + returning users searching the brand name 3,000. Clicking a link inside desktop software is usually counted as a direct visit, so part of the cc-switch traffic is placed under direct. The 5% who click through after seeing the cc-switch sponsor slot, the 3% after seeing promotional articles and the 10% after hearing community recommendations are guesses. The records warn that the 21,500 checkout page visits are exactly the same as venice.ai and other sites and may be tiered estimates from the data source; they are used as is.
Checkout: checkout.stripe.com
New-user source: Other (Direct / referral / email / paid). Direct 68.5%, referral 28.4%: developer sites and affiliates such as gitcode.com 5.52%, hvoyai.com 1.63%, oken.ai 1.51% (the top three channels include Affiliate); search terms are mostly brand keywords (68.5%), plus 'deepseek' and 'workbuddy 接入' (workbuddy integration). Existing users come straight back to top up and check usage.
Top channel measured by Similarweb, others estimated by rank
| Attributable new users, last month | Share |
|---|---|
| Other | 100% |
Direct + brand search 68%. Direct and brand search count as viral only when the site shows sharing mechanics; the largest share wins, under 5% means Other.
| Country | Share of visits |
|---|---|
| China | 58.73% |
| United States | 13.6% |
| Singapore | 8.39% |
| Taiwan | 7.44% |
| Hong Kong | 6.15% |
English and Chinese bilingual
DeviantArt, X (Twitter), YouTube
5 keywords · brand ~68% · organic 100% / paid 0%
| Similarweb top 5 keywords |
|---|
| packyapi |
| deepseek |
| packyapi.ai |
| workbuddy 接入 |
| grok |
Non-brand queries: deepseek, workbuddy 接入grok
12 referring sites
| Site | Share of referrals |
|---|---|
| gitcode.com | 5.52% |
| hvoyai.com | 1.63% |
| oken.ai | 1.51% |
| Category | Share of referrals |
|---|---|
| Programming and Developer Software | 74.78% |
| Computers Electronics and Technology - Other | 19.09% |
| Graphics Multimedia and Web Design | 0.91% |
| Beauty and Cosmetics | 0.91% |
| News & Media Publishers | 0.26% |
About 21,490 last month (Similarweb estimate), #124 of 606; the month before was 0. Opening checkout is not paying, so this is not revenue.
New-user source: Other (Direct / referral / email / paid). Mostly referrals, email or paid, or mostly direct with no clear source Direct 68.5%, referral 28.4%: developer sites and affiliates such as gitcode.com 5.52%, hvoyai.com 1.63%, oken.ai 1.51% (the top three channels include Affiliate); search terms are mostly brand keywords (68.5%), plus 'deepseek' and 'workbuddy 接入' (workbuddy integration). Existing users come straight back to top up and check usage.
Pay as you go, e.g. gpt-5.4 input $12.5 per million tokens, output $75 per million tokens (list price at 72% of official)
Dossier confidence: medium. Parts 1 and 2 come from reading the site and Similarweb data; part 3 from public reporting, founder posts, Wayback snapshots, WHOIS and similar, with speculation marked.
The number: checkout visits is Similarweb's estimate of visits from the site to a checkout payment page; opening checkout is not paying, so this is not revenue. Small sites land on fixed buckets.
Data through last month. Page updated 2026-10-31.