Inside a knowledge vault

Both the project vault and personal vault are built from four sub-vaults. Each sub-vault organizes a different type of knowledge and runs its own synthesis agent independently.

Context vault (ContextVault)

What's active in this project right now - the current focus, active files, and immediate working context. Synthesized as part of every knowledge cascade.

Pattern vault (PatternVault)

Recurring workflows and behaviors. What you do repeatedly: test patterns, naming habits, architectural preferences. Synthesized as part of every knowledge cascade.

Decision vault (DecisionVault)

Architectural choices and their rationale. Why Zod instead of Yup. Why functional components over classes. The decisions that shape everything downstream. Synthesized as part of every knowledge cascade.

Petnames vault (PetnamesVault)

Memorable session identifiers and naming patterns. Tracks the short names you use for sessions so they remain meaningful over time. Synthesized as part of every knowledge cascade.

Knowledge vault aggregate (KnowledgeVault)

After all four sub-vaults synthesize, the knowledge vault aggregates their outputs into two files:

• FAI-PROJECT.md - project-scope knowledge
• FAI-PERSONAL.md - personal-scope knowledge

These are what every connected AI reads on startup.

When synthesis runs

All four sub-vaults synthesize together as part of a knowledge cascade. The cascade triggers automatically after enough sessions have sealed - the threshold is configured when you run fai synthesize.

You never call synthesis manually. It runs when you've done enough work to make it worthwhile.