fastmemory

FastMemory

Overview

FastMemory is the horizontal layer of truth for AI agents.

It provides a structured way to store and retrieve project knowledge using the CBFDAE (Component, Block, Function, Data, Access, Event) ontology. This reduces hallucinations by ensuring the agent always references a verified, structured cognitive graph of the codebase instead of relying on transient context.

When to Use

Use FastMemory whenever you are:

• Designing a new system or feature.
• Documenting existing complex logic.
• Needing to cross-reference architectural decisions from previous sessions.
• Struggling with context limits (FastMemory provides compressed, semantic retrieval).

Symptoms that you need FastMemory:

• You keep forgetting how a specific module interacts with another.
• You find yourself "guessing" the names of data structures or events.
• The project is growing too large for the current context window.

CBFDAE Ontology

Entity	Description	Example
Component	High-level logical unit	AuthService, PaymentGateway
Block	Functional grouping within a component	LoginFlow, TokenExchanger
Function	Specific atomic operation	validateCredentials, issueJWT
Data	Data structures or schemas	UserData, AuthResult
Access	Permissions and visibility	Public, AdminOnly
Event	State changes or notifications	UserLoggedIn, AccessDenied

Implementation

1. Extraction

Use the extract_cbfdae.py tool to extract structured data from your design markdown.

python3 skills/fastmemory/extract_cbfdae.py design.md

2. Integration

When starting a task, query FastMemory for relevant context:

# Query for Auth component patterns
python3 skills/fastmemory/query_memory.py --query "Component: Auth"

Common Mistakes

• Too much detail: Don't map every line of code. Map at the architectural level.
• Stale memory: Forget to update the graph after a significant refactor.
• Flat structure: Forgetting to link Blocks to their parent Components.

Real-World Impact

• 30x Speedup: Delta updates to the cognitive graph are significantly faster than full RAG re-indexing.
• Zero Hallucination: Architecture is strictly enforced by the CBFDAE schema.