sharememory_user

Zero-train multi-user shared memory with A→C selection (A→B→C preserved). Stores to JSON; embeddings and QC have offline fallbacks.

Quickstart

Using python 3.12.

python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

# Initialize storage
PYTHONPATH=. python -m sharememory_user.cli init

# Ingest a dialog
PYTHONPATH=. python -m sharememory_user.cli ingest --user_id u1 --profile sample_data/user_u1_profile.json --dialog sample_data/dialog1.txt

# Retrieve for a user with peers
PYTHONPATH=. python -m sharememory_user.cli retrieve --user_id u1 --task "deploy X to Y and autoscale on Z" --peers sample_data/peers.json --top_k 5

# Build prompt blocks (COT + KG)
PYTHONPATH=. python -m sharememory_user.cli prompt --user_id u1 --task "deploy X to Y and autoscale on Z" --peers sample_data/peers.json --top_k 5

Embeddings: BGE-M3 by default

• Default model: BAAI/bge-m3 (SMU_EMBED_MODEL), with SMU_EMBED_USE_HF=1.
• Fallback: hashing embeddings with SMU_EMBED_DIM (default 1024) when HF not available.

Env toggles

• SMU_EMBED_USE_HF=1 to use HF; SMU_EMBED_MODEL=BAAI/bge-m3 to set model.
• SMU_EMBED_DIM to set fallback embedding dim. "# agent-share"