RecallOptions

Defined in: src/lib/memory/types.ts:82 

Extended by

• ReflectOptions

Properties

bm25AdmissionDivisor?

optional bm25AdmissionDivisor: number

Defined in: src/lib/memory/types.ts:143 

Divisor mapping BM25 scores to the admission floor. Default: 50.

budget?

optional budget: Budget

Defined in: src/lib/memory/types.ts:90 

Search depth. Default: ‘low’.

ceWeight?

optional ceWeight: number

Defined in: src/lib/memory/types.ts:129 

Multiplicative cross-encoder blend weight. Default: 0.1.

conversationId?

optional conversationId: string

Defined in: src/lib/memory/types.ts:98 

Restrict chunk search to one conversation. Chunk-only.

decomposeOptions?

optional decomposeOptions: object

Defined in: src/lib/memory/types.ts:108 

Auth + endpoint for the LLM-based query decomposition pass. Without these, decompose is skipped even at budget: 'high'. Mirrors the shape used by searchVaultMemories.

apiKey

apiKey: string

baseUrl?

optional baseUrl: string

model?

optional model: string

excludeConversationId?

optional excludeConversationId: string

Defined in: src/lib/memory/types.ts:100 

Exclude one conversation from chunk search. Chunk-only.

folderId?

optional folderId: string | null

Defined in: src/lib/memory/types.ts:96 

Vault folder filter. Vault-only.

includeChunks?

optional includeChunks: boolean

Defined in: src/lib/memory/types.ts:92 

Include source chunks for fact memories that have provenance. Default: false.

limit?

optional limit: number

Defined in: src/lib/memory/types.ts:86 

Max items returned. Default: 8.

maxTokens?

optional maxTokens: number

Defined in: src/lib/memory/types.ts:88 

Token budget for the result; greedy-truncates after ranking. Not yet enforced; reserved for W1.

minScore?

optional minScore: number

Defined in: src/lib/memory/types.ts:102 

Drop results below this score. Default: 0.1 for facts, 0.5 for chunks (mirrors today’s defaults).

mmr?

optional mmr: boolean

Defined in: src/lib/memory/types.ts:135 

Apply MMR diversification after ranking (rerank pipeline only). Default: false.

now?

optional now: number

Defined in: src/lib/memory/types.ts:120 

Reference “now” for resolving relative temporal phrases in the query (“last week”, “yesterday”, “N days ago”). Default: Date.now(). Override for back-dated evaluation harnesses (bench corpora dated 2021–2023) and for deterministic tests — otherwise the W6 lane resolves windows in 2026 and never overlaps stored event_time.

proofCountAlpha?

optional proofCountAlpha: number

Defined in: src/lib/memory/types.ts:141 

Proof-count log-boost scale. Default: 0.1.

recency?

optional recency: RecencyOptions

Defined in: src/lib/memory/types.ts:133 

Recency decay curve overrides (per-year decay slope, floor, no-date multiplier).

recencyAlpha?

optional recencyAlpha: number

Defined in: src/lib/memory/types.ts:131 

Recency boost slope in the fused ranker. Default: 1.0.

rerankTopN?

optional rerankTopN: number

Defined in: src/lib/memory/types.ts:127 

Number of candidates fed to the cross-encoder rerank stage. Default: 30.

rrfK?

optional rrfK: number

Defined in: src/lib/memory/types.ts:145 

RRF smoothing constant for lane fusion (facts × chunks and side lanes). Default: 60.

scopes?

optional scopes: string[]

Defined in: src/lib/memory/types.ts:94 

Vault scope filter. Vault-only.

supersessionBoost?

optional supersessionBoost: number

Defined in: src/lib/memory/types.ts:137 

Supersession score-gap transfer factor. Default: 0.8.

supersessionWindow?

optional supersessionWindow: number

Defined in: src/lib/memory/types.ts:139 

Hard cap on the supersession candidate window. Default: 50.

types?

optional types: MemoryKind[]

Defined in: src/lib/memory/types.ts:84 

Which kinds to search. Default: [‘fact’].