export interface VisionEntity {
    id: string;
    sourcePath: string;
    hash: string;
    capturedAt: string;
    description: string;
    ocr?: string;
    tags: string[];
    project: string;
    embedding?: number[];
}
export declare class MultimodalVision {
    private ollamaModel;
    private ollamaBaseUrl;
    private initialized;
    init(): Promise<void>;
    /** Ingest an image file. Returns the persisted VisionEntity. */
    ingestImage(filePath: string, opts?: {
        project?: string;
        tags?: string[];
        prompt?: string;
    }): Promise<VisionEntity>;
    /** Semantic search across stored image descriptions. */
    search(query: string, topK?: number, project?: string): Promise<VisionEntity[]>;
    /** Recent ingests, newest first. */
    recent(limit?: number, project?: string): Promise<VisionEntity[]>;
    /** True if a vision model was detected. */
    hasVisionModel(): boolean;
    private describe;
    private appendIndex;
    private loadAll;
}
/** OCR stub. TODO: wire up tesseract.js or system tesseract for real OCR.
 *  Returns null today; vision model + description handles text most of the time. */
export declare function ocrExtract(_filePath: string): Promise<string | null>;
export declare function getMultimodalVision(): MultimodalVision;
export declare function resetMultimodalVisionForTests(): void;
//# sourceMappingURL=multimodal-vision.d.ts.map