01
Traffic
Reach
Creators / SEO / private domain — three lanes at once
AI NATIVE · Cross-border Infrastructure Let global trade
Let global trade
flow down one chain
The AI-native infrastructure for going global. Sites, inquiries, matchmaking, traffic — one workflow, end to end.
Not four products. One closed loop. traffic → site → inbox → deal
02
Site
Sitebox
LLM runs brand → code in 30 minutes
03
Inbox
Inbox
7×24 understands, quotes, follows up
04
Match
Match
Hard constraints + cascading relax. No vector guessing.
Traffic in → site catches → engine converts → matched into orders.
Why us
Not a ranking war. Not a pile of plugins. One workflow that actually closes deals.
The job
Traditional trade
SaaS patchwork
Link4a
Traffic
Buy ads on marketplaces / hand out cards at fairs
Tools don't bring traffic
Creators / SEO / private domain, built in
Inquiries
Human-handled; lost if not answered in 3 days
Tool islands, data doesn't flow
LLM understands + replies, 7×24, real time
Matchmaking
Down to a salesperson's gut
Vector similarity; constraints fail
Hard constraints + cascading relax, explainable
Four modules, one loop
Four things, done together — not four SKUs sold apart.
Traffic comes in, the site catches it, the engine converts it, matchmaking closes it.
Sitebox
Site generation
A branded standalone site in 30 minutes
- LLM runs brand → copy → code
- Multi-template · multi-language
- Deploys to Cloudflare Pages
Inbox
Inquiry AI
7×24 understand + quote + follow up
- WhatsApp / email fully wired
- Auto-translation across languages
- Funnel state machine, visualized
Match
Matchmaking engine
Hard constraints + LLM re-rank
- Category / process / MOQ / export
- 7-level cascading relaxation
- Recommendation reasons in plain Chinese
Reach
Content engine, omni-channel traffic
One content engine — get the factory seen everywhere
- Social matrix: TikTok / Instagram / X / LinkedIn
- SEO + GEO: ranked on Google and quoted by AI engines
- Outbound + lead mining: LinkedIn / customs data / email
Inside the matchmaking engine
A database, a light model, a heavy model and deterministic tools — orchestrated into one workflow that runs.
We borrow the task-orchestration paradigm from agentic coding: each phase does one job well, and hands off explicitly.
RFQ in
Phase 1
Understand
Parse to JSON
Confidence < 0.4 → escalate to human
→
Phase 2
Pre-filter
7-level cascading relax
Candidates ≤ 100 · debuggable
→
Phase 3
LLM re-rank
Recommendation reasons
Recall-biased · metadata packing
→
Phase 4 · reserved
Agent loop
4-tool loop
Only on low scores · ≤ 8 turns
Result
We don't treat the model as a search engine, and we don't string tools into a plugin market.
Light models read intent, the database does the hard filtering, the heavy model only earns its keep on the few hard cases — every step is explainable.
De-identified RFQ
"Looking for a supplier of 304 stainless steel tumblers, double-wall vacuum, 500ml, laser logo, MOQ around 2000, ship to Germany before Q4."
MongoDB candidates · 0
YongkangJinhuaYiwu +117
Engine output · top 2
Factory A · Yongkang, Zhejiang
0.00
304 verified · vacuum line in-house · laser marking · exported to DE, has CE
Factory B · Jinhua, Zhejiang
0.00
MOQ 1500, more flexible · slightly longer lead time, relaxed at level 3
Real cases
Real factories. Real authorization. Real orders.
CASE 01 
Zhejiang · Hardware · Stainless steel
"From a fair business card to inquiry automation — first German order in 3 weeks."
Site live 10 days
First order closed 21 days
Monthly inquiries +280%
CASE 02 
Guangdong · Small appliances · OEM
"The AI handled night-time inquiries in 6 timezones — no more missed leads."
Reply latency < 2 min
Quotes / month 320+
Close rate +46%
CASE 03 
Fujian · Outdoor gear · DTC
"Matched to 4 TikTok creators in a week — the first one drove 1,100 orders."
Creators matched 4 / week
First-batch orders 1,100+
GMV growth +3.2×
0 +
Factories served
0 +
Sites launched
0 h
Avg. first-order response
0 %
Inquiry parsing accuracy
About
We believe the next decade of cross-border trade is AI-native — not a model used as a search box, not tools stitched into a plugin market, but database, light models, heavy models and deterministic tools orchestrated into a workflow that actually runs.
Link4a is making the link between Chinese factories and global buyers shorter, sharper and more explainable.
Source from verified Chinese factories.
We make the matchmaking actually work — sample to mass production.
Latest
See all articles → Latest
2026 · 05 Matchmaking engine Phase 3 live — avg match time down to 3s New
2026 · 03 Sitebox cuts brand-to-deploy to under 30 minutes
2026 · 01 Factories served crosses 300
Contact
The fastest ways to reach us
Fill the form or email us — business responds within 4 hours on workdays.
Email [email protected]
Submissions route to our ops team and get a reply within 4 working hours.
✓
Received — thank you
Our business team will reach out within 4 working hours.