Optional topKOptional typesOptional scopesOptional tagsOptional entitiesOptional minOptional usableOptional timeOptional after?: numberOptional before?: numberOptional neutralIf true, skip emotional congruence bias (useful for factual lookups).
Optional maxOptional visibility ceiling for prospective alerts. Items registered with a
tierRank greater than this are withheld from the assembled "Reminders"
section. Used only for prospective gating in assembleForPrompt;
retrieve ignores it.
Optional hydeEnable HyDE (Hypothetical Document Embedding) for memory retrieval.
When true and a HyDE retriever is configured on the memory manager,
the system generates a hypothetical memory trace matching the query
before embedding. This produces embeddings that are closer to actual
stored memories, improving recall — especially for vague or abstract
recall prompts (e.g. "that thing we discussed about deployment").
Adds one LLM call per retrieval. Use for important lookups where recall quality matters more than latency.
false
Optional policyShared retrieval profile and confidence policy.
Optional scoringOverride the 6-signal retrieval weights for this call. Missing keys fall back to DEFAULT_SCORING_WEIGHTS. Useful for ablation studies (zero one weight at a time and measure Δaccuracy) and for A/B testing alternate weight configurations without mutating global defaults.
Restrict results to traces whose trust policy permits the listed capabilities. Pass a single capability or an array (AND semantics: a trace must permit every requested capability). Applies the same staleness check as canUseFor when the policy declares
requiresReverificationAfterMs.