Module multimodal 

Multi-Modal Memory for Microscope Memory.

Extends beyond text to store and recall images (perceptual hashes), audio (spectral fingerprints), and structured data (typed key-value pairs) within the same spatial coordinate framework.

Binary format: modalities.bin (MOD1)

Modalities are stored as a sidecar index — the core BlockHeader (32B mmap’d) is unchanged. Each entry maps a block_idx to its modality metadata.

Structs

AudioMeta

ImageMeta

ModalityIndex

Sidecar index mapping block indices to modality metadata.

ModalityStats

StructuredMeta

Enums

FieldValue

Modality

Functions

compute_phash

Compute perceptual hash (dHash-like) from grayscale pixel data. Expects row-major grayscale pixels.

compute_spectral_fingerprint

Simple spectral fingerprint from audio samples. Divides into 16 frequency bands and computes energy per band.

hamming_distance

Hamming distance between two byte arrays.