ForgeMemory¶

ForgeMemory persists business rules across sessions so you don't have to re-specify them in every generation call.

Adding Rules¶

forge.memory.add_rule("user", "Users under 18 get account_type='restricted'")
forge.memory.add_rule("financial", "Max 3 active loans per customer at any time")
forge.memory.add_rule("user", "Israeli phone numbers use format 05x-xxx-xxxx")

Rules are stored in .forge/ in your project directory and injected into AI prompts automatically on every generation call.

How Rules Inject¶

Rules are re-read from disk on every create_batch() call — not cached in memory. This means:

Edit a rule file → next generation call picks it up immediately
No restart needed
Always fresh

Skeptical Memory¶

FixtureForge validates stored rules against the live schema before using them:

# If the schema changes (e.g., account_type field removed),
# the rule is flagged as potentially invalid
result = forge.memory.validate_against_schema(User)
if result.has_conflicts:
    print(result.conflicts)

Progressive Forgetting¶

FixtureForge never stores what it can re-derive:

Field names? The model schema has them.
Field types? Same.
Default values? Pydantic handles those.

Only business rules that exist nowhere else in the code are kept.

Field-name-only rules are rejected:

forge.memory.add_rule("user", "name is a string")
# Rejected: re-derivable from schema. Nothing stored.

Storage Structure¶

.forge/
  FORGE.md          # index — topic list + line counts (max 200 lines)
  user.md           # user domain rules (max 10KB)
  financial.md      # financial domain rules (max 10KB)
  .dream_state.json # ForgeDream run state
  coverage_gaps.json

Stats¶

forge.stats()
# {
#   "memory": {
#     "topics": 3,
#     "total_kb": 2.4
#   }
# }