I buy industrial equipment for a living, at ISATEK. Kanpo comes out of that job: sending enquiries to suppliers, collecting the quotes that come back, comparing them, issuing purchase orders, and chasing the ones that never arrived.
The tools I tried assumed a company shaped differently from mine: a procurement department, an implementation partner, and somebody to attend the training. Kanpo assumes one person who needs three prices.
What it is
One workspace where a company can buy and sell without switching hats.
Buying: publish an RFQ, invite suppliers, collect quotes that arrive in the same shape, compare them, award, and issue a purchase order.
Selling: publish an offer, send a link, take an acceptance or a counter-offer, and receive the purchase order into your own workspace.
Suppliers respond from a link. No account, no password, no portal tour: a magic link opens the line items and the deadline. A supplier who wants to keep the relationship can claim the invitation into their own workspace afterwards.
VAT identity is looked up, not typed. A company’s details come from the Greek tax authority and the EU registry through libraries I wrote and maintain: VatchexGreece and VatchexVies.
Built the way it is for a reason
Elixir and Phoenix LiveView, so screens update without a page reload and the whole product is one deployable application rather than a front end talking to an API somebody else owns.
It runs on a phone, installs as an app, and sends web push notifications, because the people who answer RFQs are often nowhere near a desk.
Greek and English from the first release. Every string is translated rather than hardcoded, so adding a language is an addition rather than a rewrite. The plumbing is in place; the languages come as the customers do.
MCP native, on every plan. Kanpo exposes its own MCP server, so an agent can work the same workspace a person does — inquiring about an RFQ, checking a quote, following a purchase order. Each member gets their own MCP token on every plan, including Solo. This is not an enterprise feature held back as leverage; it is where procurement is going, and a tool that cannot be driven by an agent will be driven around.
An optional REST API for integrations. Integrate is an add-on on every plan, at one price: a documented API and signed webhooks for the company whose ERP or accounting has to hear about a purchase order without anyone retyping it.
The three plans
The ladder is named after units of organisation, because “Basic” implies the least and “Pro” judges the solo trader an amateur. Every plan is one workspace: the difference is how many people work in it and how far back it remembers.
| Plan | For | People who work | People who only look |
|---|---|---|---|
| Solo | the one-person company that buys occasionally and wants one place for it | 2 | 1 |
| Team | a small company where several people raise requests and the same people approve them | 6 | 2 |
| Group | the same work with more hands and a longer memory: more seats, more catalog, more machine calls | 12 | 4 |
Readers — the accountant, the auditor, the owner who wants to see without touching — are counted and never billed. Nobody pays per seat. Storage, catalog size and the machine-call ceiling scale with the plan, and extra storage comes in flat packs rather than as a hidden tier.
The AI assist inside the app starts at Team. The MCP surface is on every plan, including Solo.
The caps are published, not discovered at the point of frustration.
The interface
Procurement tools have a reputation for looking like a tax form from 2006, and the reputation is earned. Kanpo is built to be readable: one workspace instead of a buyer persona and a supplier persona, work streams instead of modules, empty states that explain the next step, and a form a supplier can complete on a phone without instructions.
The rule is written into its own UX document: if either side feels like Coupa training, we failed.
What it refuses
Forced supplier registration. A supplier should not need an account to give you a price.
Per-seat pricing. Charging by the seat punishes the behaviour the software exists to encourage: more suppliers, more quotes.
“Contact sales.” Published prices, self-serve signup.
Implementation partners. There is nothing to implement.
The buyer/supplier persona switch. A company buys and sells. Two navbars for one business means the model is wrong.
The business model
Publisher pays, responder never. Publishing an RFQ or an offer carries a small fee; quoting or accepting costs the other party nothing. An optional success fee applies when a purchase order is created on the platform.
And the boundary: Kanpo never holds, escrows, routes or touches the money for the goods. Companies settle with each other as they do today. This is software, not a payment intermediary.
Where it stands
Pre-launch, with a founding cohort forming. The buy and sell lifecycles, the magic-link portals, the catalogue, notifications, VAT verification, MCP, the API, plans and caps exist as working code. What is left is a launch: the parts of a first release that are not features, and the support a first cohort deserves.
I built it for my own company’s purchasing first. If it holds up there, it will hold up for other companies the incumbents have decided are too small to serve.