db-seed-ai

Stop writing fake data by hand. Let AI do it.

db-seed-ai reads your SQL schema file and uses a local AI model (Ollama) to generate hundreds of rows of realistic, relationally-correct seed data — then inserts it directly into your database.

No cloud. No API keys. No cost. Runs 100% locally.

The Problem

Every developer has done this:

INSERT INTO users VALUES (1, 'Test User', 'test@test.com');
INSERT INTO users VALUES (2, 'John Doe', 'john@doe.com');
-- ... repeat 50 more times, slowly dying inside

db-seed-ai reads your actual schema, understands your foreign keys and constraints, and generates data that looks real and actually works.

Demo

# Preview what gets generated (no database needed)
$ db-seed-ai preview --schema ecommerce.sql --table users --rows 5

  Asking llama3 to generate 5 rows...

  ┌────┬───────────────────┬───────────────────────┬───────────┐
  │ id │ name              │ email                 │ role      │
  ├────┼───────────────────┼───────────────────────┼───────────┤
  │  1 │ Sarah Mitchell    │ sarah.m@gmail.com     │ admin     │
  │  2 │ James Rodriguez   │ j.rodriguez@mail.com  │ user      │
  │  3 │ Priya Patel       │ priya.p@company.io    │ user      │
  │  4 │ Marcus Chen       │ m.chen@startup.dev    │ moderator │
  │  5 │ Emma Thompson     │ ethompson@email.net   │ user      │
  └────┴───────────────────┴───────────────────────┴───────────┘

  Dry run — no data was inserted.

# Seed your actual database
$ db-seed-ai seed --schema ecommerce.sql \
    --db postgres://localhost/mydb --rows 100

  db-seed-ai v0.1.0

  Schema loaded:  6 tables
  Insert order:   categories → users → products
                  → orders → order_items → reviews

  AI model: llama3

  Generating seed data...

    Generating  categories          ✓ 10 rows
    Generating  users               ✓ 100 rows
    Generating  products            ✓ 100 rows
    Generating  orders              ✓ 100 rows
    Generating  order_items         ✓ 200 rows
    Generating  reviews             ✓ 150 rows

  Inserting into database...

    Inserting   categories          ✓ 10 inserted
    Inserting   users               ✓ 100 inserted
    Inserting   products            ✓ 100 inserted
    Inserting   orders              ✓ 100 inserted
    Inserting   order_items         ✓ 200 inserted
    Inserting   reviews             ✓ 150 inserted

  ✓ Done in 4.2s — 660 rows inserted across 6 tables

Installation

# Prerequisites
# 1. Install Go 1.22+ from https://go.dev/dl
# 2. Install Ollama from https://ollama.ai/download
# 3. Pull a model
ollama pull llama3

# Install db-seed-ai
git clone https://github.com/satyammistari/db-seed-ai
cd db-seed-ai
go build -o db-seed-ai .

Commands

preview — See data before touching your DB

db-seed-ai preview \
  --schema schema.sql \
  --table users \
  --rows 5

seed — Generate and insert

# PostgreSQL
db-seed-ai seed \
  --schema schema.sql \
  --db "postgres://localhost/mydb" \
  --rows 100

# SQLite (easiest for local dev)
db-seed-ai seed \
  --schema schema.sql \
  --db sqlite:./dev.db \
  --rows 50

# Dry run — generate but do not insert
db-seed-ai seed \
  --schema schema.sql \
  --dry-run \
  --rows 10

# One specific table only
db-seed-ai seed \
  --schema schema.sql \
  --db sqlite:./dev.db \
  --table users \
  --rows 25

validate — Check data quality

db-seed-ai validate \
  --schema schema.sql \
  --rows 10

Flags

Flag	Default	Description
--schema	required	Path to your .sql schema file
--db	required	Database connection string
--rows	100	Rows to generate per table
--table	all tables	Only seed this one table
--model	llama3	Ollama model to use
--style	realistic	realistic, minimal, edge-cases
--batch-size	500	Rows per INSERT batch
--dry-run	false	Generate but do not insert

What It Understands

db-seed-ai parses your schema and passes this context to the AI so it generates valid data:

• Foreign keys — order.user_id always references a real user that was already inserted
• CHECK constraints — status only ever gets values from ('pending', 'paid', 'shipped')
• NOT NULL — required columns are never empty
• UNIQUE — emails, slugs, and usernames never repeat
• Data types — prices are decimals, dates are timestamps, IDs are integers

Data Styles

realistic (default) — Names, emails, and text that looks like it came from a real app. Good for demos.

minimal — Short values, ASCII only, no special characters. Good for unit tests.

edge-cases — Includes NULL values, near-max-length strings, boundary numbers, and special characters. Good for QA testing.

Architecture

Six components, one job each:

• cmd/ CLI commands (seed, preview, validate)
• schema/ Parses your SQL file into Go structs
• generator/ Builds AI prompt, calls Ollama, parses JSON
• inserter/ Writes rows to Postgres or SQLite
• validator/ Checks rows against your constraints
• reporter/ Colored terminal output and progress

Trade-offs

1. Topological sort for insert order Foreign keys mean you can't insert orders before users exist. We build a dependency graph from your schema and do a DFS topological sort. Categories go in first, products second (they reference categories), orders third (they reference users), and so on. Circular FKs are detected and reported as errors.

2. Transactions per batch Every batch of 500 rows is wrapped in a BEGIN/COMMIT. If one batch fails, only that batch rolls back — not the entire run. This means a 10,000 row seed that fails at row 8,000 doesn't waste the first 7,999 rows.

3. Fetch existing IDs before generating Before asking the AI to generate rows for orders, we run SELECT id FROM users LIMIT 1000 and pass those real IDs to the AI. This is what makes FK relationships actually work instead of generating IDs that don't exist.

Contributing

go mod tidy
go test ./...
go build .

License

MIT