mem0

mem0 — Hermes Shared Memory

Query and write to the shared mem0 qdrant store (hermes_mem0, 127.0.0.1:6333). All agents (Claude, Codex, Hermes) share this store under user_id=$USER.

Quick reference

# Search
python3 ~/.hermes/scripts/mem0_shared_client.py search "<query>"

# Add a fact (with LLM extraction)
python3 ~/.hermes/scripts/mem0_shared_client.py add "<text>"

# Add verbatim (no LLM, guaranteed write)
python3 ~/.hermes/scripts/mem0_shared_client.py add "<text>" --no-infer

# Stats
python3 ~/.hermes/scripts/mem0_shared_client.py stats

Script path: ~/.hermes/scripts/mem0_shared_client.py Config: reads ~/.hermes/config.yaml (override with HERMES_CONFIG_PATH), looking for plugins.entries.hermes-mem0.config.oss. That block is OPTIONAL — if absent, the script falls back to working defaults (ollama nomic-embed-text embedder + gemma2:2b LLM, qdrant 127.0.0.1:6333, collection hermes_mem0). There is NO ~/.hermes/hermes.json; do not chase it as a missing file. Also callable from any repo worktree: python3 scripts/mem0_shared_client.py <cmd>

Store status

• Collection: hermes_mem0
• Host: 127.0.0.1:6333 (qdrant docker, Hermes-managed storage)
• Embedder: ~/.hermes/config.yaml (hermes-mem0 plugin oss block) or fallback default nomic-embed-text
• LLM for extraction: ~/.hermes/config.yaml (hermes-mem0 plugin oss block) or fallback default gemma2:2b

Ingestion (backfill)

The extractor at scripts/mem0_extract_facts.py scans three sources:

Source	Path	Agent ID
Hermes	~/.hermes/	hermes
Claude Code	~/.claude/projects/	claude
Codex	~/.codex/sessions/	codex

Run a backfill (last 1 year = 525960 minutes):

cd ~/.hermes
python3 scripts/mem0_extract_facts.py --since 525960

Run incremental (last 65 min, for cron):

cd ~/.hermes
python3 scripts/mem0_extract_facts.py --since 65

Check state:

cat ~/.hermes/memory/extraction-state.json | python3 -m json.tool | head -20

Current ingestion status

Use python3 ~/.hermes/scripts/mem0_shared_client.py stats for live status.

When to use this skill

Use /mem0 when the user asks to:

• Search memory for past decisions, branch names, bead IDs, preferences
• Add a new fact that should persist across agent sessions
• Check ingestion status or trigger a backfill
• Debug why an agent doesn't remember something