Skip to main content

Lifecycle

Endpoint reference

For verifying ingestion status, see POST /ingestion/verify_processing. For retrieving memories, see POST /recall/recall_preferences.

Typical call sequence

For storing a new memory:
For removing a memory:

Memories vs knowledge

HydraDB distinguishes two classes of ingested content: Both go through the same processing pipeline (queued → processing → completed) and are searchable via the boolean_recall endpoint.

Key concepts

Memory – A single unit of user-specific context. Identified by source_id. Created by POST /memories/add_memory. Inference (infer) – When true, HydraDB extracts implicit signals (preferences, entities, relationships) from the content. When false (default), content is indexed as-is. Conversation pairs – User-assistant exchanges stored as memory. Useful for capturing chat history with full context. Expiry – Optional TTL in seconds (expiry_time). After this duration, the memory is automatically deleted. Useful for ephemeral context. Sub-tenants – Memories are scoped to a sub-tenant. For B2C apps, each end user typically has their own sub-tenant.