Skip to content

CI vs Dev Mode

FixtureForge has two modes of operation designed for different stages of the development lifecycle.

Dev Mode (default)

forge = Forge()
# Auto-detects provider from environment variables
  • Uses AI for semantic fields
  • Realistic, context-aware output
  • Slightly slower (AI latency ~1-2s per batch)
  • Ideal for: writing new tests, exploring edge cases, seeding dev databases

CI Mode

forge = Forge(use_ai=False, seed=42)
  • Zero network calls
  • Faker + structural generators only
  • Same seed = identical output on every machine, every run
  • Ideal for: test pipelines, snapshot tests, reproducible bugs

Use seed= in CI

Set seed= from an environment variable so you can override it locally without changing code: 

import os
forge = Forge(use_ai=False, seed=int(os.getenv("FORGE_SEED", "42")))

Seed Determinism

The seed= parameter controls:

  1. Faker — faker.seed_instance(seed) (instance-level, no global state pollution)
  2. Random — random.Random(seed) per Forge instance (fully isolated)

Two Forge(seed=42) instances produce identical data without interfering with each other.

forge_a = Forge(use_ai=False, seed=42)
forge_b = Forge(use_ai=False, seed=42)

users_a = forge_a.create_batch(User, count=5)
users_b = forge_b.create_batch(User, count=5)

assert users_a == users_b  # always True

Large Datasets

For very large datasets where AI cost would be prohibitive:

# seed_ratio=0.01 means AI generates 1% of records,
# the rest are interpolated deterministically
forge.create_large(Order, count=100_000, seed_ratio=0.01)
# Cost: ~1,000 AI records. Delivered: 100,000 records.

Streaming Mode

For datasets too large to hold in memory:

for record in forge.create_stream(User, count=1_000_000, filename="users.json"):
    pass  # process one record at a time — never loads all into RAM

Supports .json, .csv, .sql output formats.