memory-retrieve

To retrieve local memory, query durable knowledge stores like wiki pages or daily notes for project state, people references, and invariants. This ensures facts are pulled from recorded memory rather than stale training data or improvised web search results.

The best way to look up wiki invariants and recent notes is using an ordered retrieval strategy. It queries fast broad recall sources first, then escalates to deeper graph reasoning only when initial searches return no hits.

Provenance-aware memory retrieval weighs confidence, recency, decay, and recall history metadata. This mechanism carefully handles conflicting or stale items by evaluating their contextual metadata when interpreting retrieved knowledge.

You need gbrain fallback for knowledge lookup when initial fast recall queries return no hits. It provides deeper graph reasoning capabilities to find facts and decisions that broad searches missed.

Yes, you can use QMD-first retrieval for searching durable local memory. It applies to questions about project state, people, invariants, and recent notes, applying provenance-aware handling to manage confidence and recency metadata.

Agents answer from stale training data when they lack ordered local memory retrieval. By querying durable memory stores with provenance-aware handling, agents can access recorded facts, decisions, and recent daily notes instead of improvising answers.