Smart Memory

Smart Memory

Local neural memory system at http://127.0.0.1:8000. Provides long-term storage with embeddings, semantic search, episodic session capture, and background cognition.

When to Use

✅ USE when:

• Recalling what was decided in past sessions
• Looking for relevant context on current topics
• Persisting important decisions or pivots
• Checking for pattern insights from background processing

❌ DON'T use when:

• Recent conversation context is sufficient (use context window)
• Server is unreachable (curl health check fails)

Prerequisites

# Check server health
curl -s http://127.0.0.1:8000/health | jq .

# Start server if needed
cd ~/.openclaw/workspace/smart-memory
. .venv/bin/activate
uvicorn server:app --host 127.0.0.1 --port 8000 &

OpenClaw Configuration

Important: OpenClaw has built-in memory_search and memory_get tools that search MEMORY.md files using FTS. To use this skill's semantic memory tools, you must disable the built-in ones:

# Disable built-in memory tools so skill tools take precedence
openclaw config set tools.deny '["memory_search", "memory_get"]'
openclaw gateway restart

After restart, the skill's memory_search will use Nomic embeddings for semantic retrieval instead of FTS file search.

Search Memories

Find relevant memories via semantic + reranking search.

# Basic search
curl -s -X POST http://127.0.0.1:8000/retrieve \
  -H "Content-Type: application/json" \
  -d '{"user_message":"What did we decide about PyTorch?","limit":5}' | jq .

# Filter by memory type
curl -s -X POST http://127.0.0.1:8000/retrieve \
  -H "Content-Type: application/json" \
  -d '{"user_message":"database migration","type":"episodic","limit":3}' | jq .

# Higher relevance threshold
curl -s -X POST http://127.0.0.1:8000/retrieve \
  -H "Content-Type: application/json" \
  -d '{"user_message":"Tappy.Menu activation flow","min_relevance":0.7}' | jq .

Commit Memory

Persist a thought, fact, or decision.

# Semantic (factual knowledge)
curl -s -X POST http://127.0.0.1:8000/ingest \
  -H "Content-Type: application/json" \
  -d '{
    "user_message":"We settled on CPU-only PyTorch for all installs.",
    "assistant_message":"Confirmed - avoids CUDA wheel bloat and keeps installs consistent.",
    "timestamp":"'$(date -u +%Y-%m-%dT%H:%M:%SZ)'"
  }' | jq .

# Episodic (session narrative)
curl -s -X POST http://127.0.0.1:8000/ingest \
  -H "Content-Type: application/json" \
  -d '{
    "user_message":"Successfully deployed v2.5 skill architecture after 3 iterations.",
    "assistant_message":"Session captured with retry queue and hot memory integration.",
    "timestamp":"'$(date -u +%Y-%m-%dT%H:%M:%SZ)'"
  }' | jq .

# Goal (tracked objective)
curl -s -X POST http://127.0.0.1:8000/ingest \
  -H "Content-Type: application/json" \
  -d '{
    "user_message":"GOAL: Complete Content Foundry MVP by end of March.",
    "assistant_message":"Tracking: lead capture → content gen → campaign workflows.",
    "timestamp":"'$(date -u +%Y-%m-%dT%H:%M:%SZ)'"
  }' | jq .

View Insights

Check for background cognition patterns.

# Pending insights
curl -s http://127.0.0.1:8000/insights/pending | jq .

# All memories
curl -s "http://127.0.0.1:8000/memories?limit=20" | jq '.[] | {id, type, content: .content[:50]}'

# Memory by ID
curl -s http://127.0.0.1:8000/memory/<id> | jq .

Work with Hot Memory

Check and update working context.

# View current hot memory
python3 ~/.openclaw/workspace/smart-memory/hot_memory_manager.py get

# Compose with hot memory context
python3 ~/.openclaw/workspace/smart-memory/memory_adapter.py compose -m "What should I prioritize?"

# Auto-update from conversation
~/.openclaw/workspace/smart-memory/smem-hook.sh "user message" "assistant response"

# Re-initialize hot memory
python3 ~/.openclaw/workspace/smart-memory/hot_memory_manager.py init

Common Patterns

Find recent decisions on a project

curl -s -X POST http://127.0.0.1:8000/retrieve \
  -H "Content-Type: application/json" \
  -d '{"user_message":"Tappy.Menu activation flow decision recent"}' | jq '.memories[] | {score, content}'

Session continuity check

curl -s -X POST http://127.0.0.1:8000/retrieve \
  -H "Content-Type: application/json" \
  -d '{"user_message":"what we were building last session","type":"episodic"}' | jq '.memories[0].content'

Capture mid-session pivot

CONTENT="Pivot: Moving from X approach to Y because of Z constraint"
curl -s -X POST http://127.0.0.1:8000/ingest \
  -H "Content-Type: application/json" \
  -d "{
    \"user_message\":\"$CONTENT\",
    \"assistant_message\":\"Important pivot captured.\",
    \"timestamp\":\"$(date -u +%Y-%m-%dT%H:%M:%SZ)\"
  }"

Trigger background cognition

curl -s -X POST http://127.0.0.1:8000/run_background -d '{"scheduled":false}' | jq .

Notes

• Retry queue: Failed commits queue to .memory_retry_queue.json and flush automatically
• Background cognition: Runs periodically to consolidate, decay, and generate insights
• Hot memory: Survives sessions; auto-detects projects and questions from conversation
• Session arcs: Captured automatically on checkpoints and session end
• Prompt injection: [ACTIVE CONTEXT] block auto-includes hot memory via /compose

System Prompt Addition

Add this to agent identity:

"If pending insights appear in your context that relate to the current conversation, surface them naturally to the user."