[Question] simhash deduplication ignores user_id and causes cross-user interference

[syy41151]

Hi, I’m currently using openmemory-py 1.3.1 and encountered some unexpected behavior in the add_hsg_memory function.
My observation: In the current implementation:

`async def add_hsg_memory(content: str, tags: Optional[str] = None, metadata: Any = None, user_id: Optional[str] = None) -> Dict[str, Any]:
simhash = compute_simhash(content)
existing = db.fetchone("SELECT * FROM memories WHERE simhash=? ORDER BY salience DESC LIMIT 1", (simhash,))

if existing and hamming_dist(simhash, existing["simhash"]) <= 3:
    print('deduplicated')
    now = int(time.time()*1000)
    boost = min(1.0, (existing["salience"] or 0) + 0.15)
    db.execute("UPDATE memories SET last_seen_at=?, salience=?, updated_at=? WHERE id=?", (now, boost, now, existing["id"]))
    db.commit()
    return {
        "id": existing["id"],
        "primary_sector": existing["primary_sector"],
        "sectors": [existing["primary_sector"]],
        "deduplicated": True
    }`

It seems that the deduplication check (SELECT * FROM memories WHERE simhash=? ...) does not filter by user_id.
This means that if two different users have similar content, they will be treated as duplicates, and salience will be boosted globally.
This can cause cross-user interference and unexpected memory updates.

Could we make the hamming_dist threshold configurable?
Currently it’s hardcoded as <= 3, but in my use case, this causes many entries to be skipped and not updated.
A configurable threshold would allow more flexible deduplication.

When using OpenAI embeddings, could we allow passing custom extra_headers to the API call?
This would be useful for scenarios where we need to include additional authentication or metadata in the request.