Admissions teams answer the same enquiry on three apps and lose track of who replied. Admito puts WhatsApp, Instagram and Facebook Messenger conversations in one inbox and follows each student from first message to admission. Connecting a channel is meant to take one tap, not an afternoon in Meta’s developer settings; that one-tap setup is what I am building now.
It is early. The first week produced the skeleton: tenant isolation, background jobs, the API and the web shell. This page describes how it is built, and will grow as each slice lands.
One program, one database
The backend is a single Go program that runs in three modes: admito api, admito worker and admito migrate. All state lives in PostgreSQL 17, including the job queue. There is no Redis, no message broker and no second service to keep in step.
Inside the program, the code is split into modules such as tenancy, channels, messaging and billing. Each module owns its tables and exposes a small Go API to the others.
Keeping institutions apart
Every institution’s data sits in the same tables, so the database itself has to refuse to mix them up. Three things make that hold:
- Three database roles. One owns the schema and runs migrations. The running app connects as a second role that owns nothing and cannot bypass row-level security. The app refuses to start as anything stronger.
- A policy on every tenant table. Each row carries an institution id, and a policy compares it with a setting that lasts for exactly one transaction. A pooled connection never carries one institution into the next request. With no setting, no rows match.
- Tests that read the real schema. One test proves two institutions cannot read or write each other’s rows. Another runs after all migrations and fails if any table lacks its policy. A build script fails if code outside the database package touches the raw connection pool.
Webhooks: save first, process later
When Meta sends a message, the API does three things and nothing else: check the signature, save the raw event and queue a job in one transaction, and reply 200. A worker picks the job up afterwards. If the worker crashes, the event is still in the database and the job runs again.
What is next
The first real slice takes in messages from Meta and shows them in the inbox. After that come sign-in, the student pipeline, AI drafting and billing, in that order.