veriseed

Empty database → believable, working data. One command. Veriseed reads your Prisma schema and writes foreign-key-correct rows — deterministically: --seed 42 (the default) regenerates identical data on every run, from your current schema, so CI seeds never rot; --rows scales it for perf testing. Runs locally: no API key, no account, works offline once installed. Zero dependencies.

npx veriseed demo                                      # built-in sample schema → SQL (try this first)
npx veriseed seed --schema ./prisma/schema.prisma --rows 20 --out seed.sql

What you get (npx veriseed demo):

BEGIN;
INSERT INTO "User"    ("id","email","name","city","country") VALUES (2,'[email protected]','Aoi Suzuki','Fukuoka','Japan');
INSERT INTO "Post"    ("id","title","authorId")              VALUES (3,'Local field engine',2);  -- authorId → a User that exists
INSERT INTO "Comment" ("id","body","postId")                 VALUES (1,'Nice graph.',3);          -- postId  → a Post that exists
COMMIT;
-- name+city+country agree within the row; every foreign key points at a real parent. Reproducible via --seed.

Honest status (building in public). This is the free, open-source core. It produces FK-correct rows with light locale coherence (a name, city, and country that match within a row). The Pro layer — genuinely AI-realistic, coherent values via a local model, plus masking and subsetting — is in the works. No cloud, ever. Waitlist / founding pre-order: https://veriseed.netlify.app.

Why

Seed data is a treadmill: your app needs data that matches your schema, and your schema keeps changing. Every current fix fails somewhere:

• Hand-written seeds / fixtures encode yesterday's schema — they break on every migration and make CI flaky, because "roughly the same data" isn't the same data.
• Asking an AI chat is genuinely fine for a one-off small sample (and its values are more lifelike than this Core's — honestly). But it truncates past dozens of rows, invents parent IDs that don't exist, returns different data every ask, and can't run unattended in CI.
• Faker runs locally but has no schema awareness — no FK ordering, no cross-column coherence. You end up hand-writing the toposort, the FK-value-picking, and per-row coherence, with a fresh script for every schema.
• supabase-community/seed (the open-sourced Snaplet) is free and OSS, but effectively stagnant.
• You can't copy prod into dev/CI for privacy/compliance — so you're generating fake data anyway.

veriseed regenerates the data from the schema, deterministically: parents before children, every FK picked from parent rows that actually exist, byte-identical output for the same --seed — so the seed can live in your repo as a command instead of a fixture file. Any --rows count for scale testing. And it all runs on your machine: nothing about your schema or data is transmitted, to me or to anyone else. The Core does this today; AI-realistic values via a local model, PII masking, and subsetting are the Pro roadmap (see Honest status above).

What the core does

• Parses your Prisma schema (@id, @relation, @map/@@map, enums, optionals, @default).
• Topologically orders tables so foreign keys always resolve.
• Generates rows with light locale coherence (name + city + country agree within a row).
• Picks FK values from already-generated parent PKs → no orphans.
• Emits a single transactional SQL file (BEGIN; … COMMIT;). Columns with a @default (e.g. createdAt) are left for the DB to fill.

Usage

npx veriseed demo --rows 5                 # see it work on a User/Post/Comment sample
npx veriseed seed --schema schema.prisma   # seed from your schema (default 10 rows/table)

# options
#   --rows <n>    rows per table (default 10)
#   --seed <n>    PRNG seed for reproducible output (default 42)
#   --out <file>  write to a file instead of stdout

Roadmap (founding members help decide the order)

• Drizzle schema + live Postgres introspection (--db <url>)
• Pro: local-LLM coherent values (real addresses, valid-looking IDs, semantically consistent rows), PII masking, prod-clone subsetting
• More databases / ORMs

License

MIT.