Typed events from a Prisma schema

@walcast/typegen-prisma generates TypeScript row types for your walcast events from a Prisma schema. Pure text-in, text-out codegen: no Prisma runtime, no walcast runtime — the generated file is self-contained, with zero imports.

npx @walcast/typegen-prisma --schema prisma/schema.prisma --out src/walcast-types.ts

Output: wrote src/walcast-types.ts: 4 table types, 1 enum. Re-run whenever the schema changes (a postinstall or generate script is the usual home).

Using the generated types

import { Walcast } from 'walcast'
import { isChange } from './walcast-types'

const tr = new Walcast({ connection: process.env.DATABASE_URL! })

for await (const event of tr.changes()) {
  if (isChange(event, 'users')) {
    // event is now WalcastEvent<'users'>:
    event.after?.email // string
    event.after?.created_at // string — Postgres text timestamp
    event.before // UserRow | null
  }
  tr.ack(event)
}

isChange(event, table) is a type-guard that narrows on event.table. The generated file also exports:

• <Model>Row — one interface per model (optional fields become | null)
• WalcastTables — table name → row type map (respecting @@map)
• WalcastEvent<T> — a ChangeEvent narrowed to table T
• One union type per Prisma enum

What maps to what — and why

Types are honest about what pgoutput actually delivers: text-format tuples, decoded conservatively. The generated types describe the events as they arrive, not as Prisma's client would hydrate them.

Prisma type	Generated TS type	Why
String	string
Boolean	boolean	decoded from t/f
Int	number	int2/int4 fit safely
Float	number	float4/float8
BigInt	string	int8 can exceed Number.MAX_SAFE_INTEGER; walcast refuses to corrupt big ids silently
Decimal	string	numeric loses precision as a float
DateTime	string	Postgres text form, e.g. "2026-07-19 12:00:00+00" — parse it yourself (new Date(...)) when you're sure of the column's timezone semantics
Json	JsonValue	json/jsonb are parsed by the decoder
Bytes	string	Postgres hex text form
any T[] list	string	Postgres arrays arrive as array literal text like "{a,b}" — not decoded
enums	union of literals

Also respected: @map (column names as they appear in events), @@map (table names), optional fields (? → \| null), and relation fields are skipped (they aren't columns).

Limits

The parser is a regex pass over the schema file, not the Prisma engine. It covers models, enums, scalars, lists, optionals, and map attributes; exotic schema features (composite types, multi-file schemas, views) are not handled. Unknown scalars fall back to string — which, given text-format tuples, is also the truthful default.