An AI self-evolving system — the system learns automatically and adjusts itself for the next round.
TokoAI(51Toko) is a self-evolving AI team providing an AI find-suppliers service — the tech timeline below is the capability base behind our free find-suppliers service for overseas buyers.
Each round of improvement tackles one concrete matching scenario, and the system gets more accurate with every request it processes. The latest round is at the top.
Our global pool of reachable companies crossed 1,000,000+, including 810,000+ overseas importers/buyers and 200,000+ China-based suppliers — the first time China reachable suppliers crossed 200,000. Recent intake of major industrial shows — Aluminium China, SinoCorrugated — plus the global buyer directory, customs import records, overseas trade shows and EU procurement awards means a large share of companies now carry reachable contact details. A larger, directly-reachable pool keeps the flywheel spinning faster.
Data enrichment capability made explicit and opened to clients: first use our own high-quality company pool, then our self-built 40M+ company data foundation, then cross-web multi-source cascading enrichment. 300K+ companies enriched with verified contacts. Even a single company name can be enriched from zero. This completes our three services: a high-quality company pool, data enrichment, and AI company-matching.
488K UK importers newly ingested (about 36% already carry directly reachable contacts via enrichment). Total company pool passed 1.335M. The enrichment engine keeps running - auto-filling email/website for companies missing contacts, turning records into reachable: reachable companies climbed from 366K to 757K+, nearly doubling. The bigger the pool, the more enrichment hits - the flywheel spins faster.
Canton Fair 137+138 buyer lists ingested after contact enrichment (enrich). Total company pool passed 839k; reachable companies reached 553k (65.9% of pool). The enrichment engine keeps running, turning contact-missing companies into reachable ones continuously.
Completed contact enrichment across all data sources. Overseas reachable companies from 336K to 366K+, China from 39.8K to 40K+, adding about 50K newly reachable companies. Every company now has an email or website for direct outreach.
Our self-built AI matching engine went live. Enter a product and the system matches three types at once: overseas buyers, peer factories, and upstream suppliers - all three directions in one run. Match precision keeps iterating; smarter the more it is used.
US (USASpending, 27k), UK (FTS, 38k) and Australia (AusTender, 5.8k) government procurement data ingested. Total company pool reached 657.1k; reachable companies 359.6k.
A full year of EU TED award data - 246k awarded companies - ingested. Total pool reached 593.8k; overseas reachable companies grew from 137.2k to 357.1k. Every company carries email, industry classification, and award description.
Continuous import from multiple global trade-show sources grew the total pool from 167k to 341.4k companies (China + overseas). A 40M+ company database cross-validates and enriches the companies.
You tell us what content works and which companies are worth pursuing; the system learns it. Your feedback becomes the system's capability directly - not tossed into a suggestion box.
From a wide pool it auto-selects the companies most likely to respond, paired with the best-fitting content version, ranked by interest signals. You do not need to make the choices.
We paused to ask what users actually need. When you treat the other side as a partner rather than just another name on a list, everything changes.
Multiple viewpoints cross-verify the same signal - we only advance when it is genuinely interesting. No wasted time on a misjudgment, and no real opportunity missed.
Who opened, who clicked, who came back for a second read - the system identifies these signals automatically. No guessing which company is worth pursuing.
You say what kind of company you want; the system filters the right ones from global purchase data - each with company name, contact, and background.
What you care about is not whether the email is well written, but which company fits your offering. Now the data is right in front of you - which companies, what background.
Same offering, different words for different companies. The system keeps the versions that work and drops the rest - you do not write each message yourself.
Who to send to, when, with what subject line, and what CTA - cold emails no longer sink into the void. The system has already tested channels, send cadence, subject-line approaches, and CTA best practices for you.
TokoAI did not appear from nowhere — it runs on the AI infrastructure our team has built since 2022. Here is the key R&D that powers 51toko (latest first).
The tech accumulation lands in the "company-matching" scenario as the team's flagship product line.
"Email AI to do the work, send at night and get results in the morning" — remotely controlled and unattended, first used for deep email-attachment processing, the same interaction paradigm as TokoAI's "email to reach companies".
Decentralized cross-device multi-agent protocol — the technical source of TokoAI's self-evolving core.
Modular, plug-and-play AI tools, plus a private-data smart database/knowledge base AIDB (RAG + multi-model coordination) — the engineering base for TokoAI's tool matrix.
Self-built sales system with multiple sales modes, auto-generated script library, and follow-up — the direct predecessor of TokoAI's "AI runs your outreach".
Full AI SaaS productization with quantified delivery (10x+ measured efficiency gain), proving the "deliver by result" loop.
Operating graph databases in natural language — the same relational-mining tech behind TokoAI's matching engine.
Built text LLM Tulier-instruct (Qwen2-7B + LoRA + int8) and on-device multimodal LLM Safer-V (MiniCPM) — both beat their base models on their tasks, mastering the full stack from data construction to fine-tuning and quantization.
Deployed Llama3/Qwen series and BGE embedding models on Alibaba Cloud, Huawei Cloud, and Tianyi Cloud — and consolidated it into a fully private deployment toolchain (ModelChain), the foundation for keeping client data in-domain and for delivery.
Founded an LLM R&D studio and began the full-stack exploration from dataset construction to model fine-tuning and private deployment — the root of TokoAI's tech accumulation.
The matching engine is improved continuously with every request we process — category mapping, signal classification, and verification are tuned from real outreach feedback, not from a fixed release schedule. The more requests we handle, the more accurately we can surface suppliers who fit yours.
Our reachable company pool keeps expanding — new trade-show, customs, and public-procurement sources are integrated and cross-checked, and more suppliers become reachable as enrichment continues. Current reach: 1.61M+ total, 1.01M+ (a 63.0% share), with 815K+ overseas and 201K+ China-based reachable.
Related: Meet the TokoAI team · Back to home
Want suppliers matched to your request? Tell us what you want to source.
Submit your request →