import type { MemoryStore } from '../store/index.js';
import type { MemoryPipeline } from './types.js';
import type { Embedder } from '../embedding/types.js';
export interface SemanticPipelineConfig {
    /** Vector-capable store. Must implement `search()`. */
    readonly store: MemoryStore;
    /** Embedder used for both write-side indexing and read-side query. */
    readonly embedder: Embedder;
    /**
     * Stable id for the embedder — attached to written entries and used
     * as a filter at read time so a later embedder swap doesn't produce
     * cross-model similarity pollution. Example: `"openai-text-embedding-3-small"`.
     */
    readonly embedderId?: string;
    /** Top-k entries to consider per turn. Default 20; picker narrows further. */
    readonly k?: number;
    /** Cosine threshold below which matches are dropped. Default none. */
    readonly minScore?: number;
    /** Tier filter for retrieval. */
    readonly tiers?: ReadonlyArray<'hot' | 'warm' | 'cold'>;
    /** Tier to tag writes with. */
    readonly writeTier?: 'hot' | 'warm' | 'cold';
    /** TTL for written entries (ms from write time). */
    readonly writeTtlMs?: number;
    /** Forwarded to `pickByBudget`. */
    readonly reserveTokens?: number;
    readonly minimumTokens?: number;
    readonly maxEntries?: number;
    /** Forwarded to `formatDefault`. */
    readonly formatHeader?: string;
    readonly formatFooter?: string;
}
/**
 * Build the semantic read + write pipelines sharing a single store.
 * Returns `{ read, write }` ready to pass to `Agent.memory()` via the appropriate `defineMemory` config (or used directly via `mountMemoryRead`/`mountMemoryWrite`).
 */
export declare function semanticPipeline(config: SemanticPipelineConfig): MemoryPipeline;