Hanzo Memory

Hanzo Memory is a searchable, editable long-term memory in one place. Store facts and context as you go, then retrieve them by meaning — it powers personal recall for agents and applications through semantic vector search.

Remember

Store a memory for a user. Each memory is scoped to a userId and can carry an importance score (0-10) and arbitrary metadata.

curl -X POST https://api.hanzo.ai/v1/remember \
  -H "Authorization: Bearer hk-..." \
  -H "Content-Type: application/json" \
  -d '{
    "userId": "u_alice",
    "content": "Prefers dark mode and terse answers.",
    "importance": 7,
    "metadata": { "source": "chat" }
  }'

Search and Edit

Retrieve memories by semantic similarity, then get, update, or delete them.

# Semantic search
curl -X POST https://api.hanzo.ai/v1/memories/search \
  -H "Authorization: Bearer hk-..." \
  -d '{ "userId": "u_alice", "query": "UI preferences", "limit": 5 }'

# Delete a memory (or all memories for a user)
curl -X DELETE https://api.hanzo.ai/v1/memories \
  -H "Authorization: Bearer hk-..." \
  -d '{ "userId": "u_alice", "memoryId": "mem_123" }'

Results are ranked by relevance and importance, so the most useful context surfaces first.

Features

• Semantic search over stored memories using vector embeddings
• Multi-tenant isolation per userId
• Importance scoring to weight what matters
• Structured knowledge base of facts with confidence scores (/v1/kb/facts/*)
• Optional PII stripping and LLM-based filtering before storage
• Metadata on every memory for filtering and provenance

Memory builds on the same embedding and vector primitives as the rest of the platform — see Embeddings and Vector to build custom retrieval on top.

Related

• Embeddings — the vectors behind semantic recall
• Vector — the managed vector database for custom RAG
• MCP — expose memory to agents as a tool
• API Keys — authenticate with hk- keys
• API Reference — every endpoint at api.hanzo.ai