microscope-memory 1.2.0

Pure binary cognitive memory engine with REST API bridge for LLM integration. Zero-JSON core, mmap-based, hierarchical, sub-microsecond search.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33

"""Microscope Memory embedding server — called as subprocess for batch embedding."""
import json, struct, sys, os
from sentence_transformers import SentenceTransformer

def main():
    model_name = sys.argv[1] if len(sys.argv) > 1 else "all-MiniLM-L6-v2"
    model = SentenceTransformer(model_name)
    dim = model.get_sentence_embedding_dimension()
    
    # Write header: dimension as 4 bytes (u32)
    sys.stdout.buffer.write(struct.pack("<I", dim))
    sys.stdout.buffer.flush()
    
    for line in sys.stdin:
        line = line.strip()
        if not line:
            # Send zero vector to keep protocol in sync
            sys.stdout.buffer.write(struct.pack(f"<{dim}f", *([0.0] * dim)))
            sys.stdout.buffer.flush()
            continue
        try:
            emb = model.encode(line)
            sys.stdout.buffer.write(struct.pack(f"<{dim}f", *emb.tolist()))
            sys.stdout.buffer.flush()
        except Exception as e:
            # Send zero vector on error — Rust code filters zero vectors
            sys.stderr.write(f"embed error: {e}\n")
            sys.stderr.flush()
            sys.stdout.buffer.write(struct.pack(f"<{dim}f", *([0.0] * dim)))
            sys.stdout.buffer.flush()

if __name__ == "__main__":
    main()