Persistent Memory

The AI remembers users across conversations. Preferences, context, and important details are persisted in PostgreSQL and retrieved automatically — so users never start from zero.

Chatbot Memories dashboard showing stored memories, search filters, and management controls

The memory dashboard at /chatbot-memories — search, filter, create, and manage stored memories.

AI-driven memory persistence

The chatbot automatically saves important facts, preferences, and context during conversations using a built-in memory tool with 5 modes.

  • 5 tool modes: search, create, update, delete, list
  • AI decides what's worth remembering based on conversation context
  • Memories checked at the start of every new conversation
  • Users can also manage memories manually via the dashboard

Full-text search

Memories are searchable using PostgreSQL's built-in full-text search with GIN indexing and ts_rank relevance scoring.

  • PostgreSQL search_memory function with ts_rank scoring
  • GIN index on the memory table for fast lookups
  • Debounced search in the dashboard UI
  • The AI automatically searches memories when relevant to the conversation

User-scoped isolation

Memories are isolated per user. When a user is in a conversation, the AI only sees and stores memories scoped to that user. Personal memories are used when not in a conversation.

  • Automatic scoping based on active user context
  • Clean separation between personal and user-level memories
  • No cross-user data leakage by design
  • Session auto-assignment handles user context automatically

Memory management dashboard

A full dashboard UI at /chatbot-memories for viewing, searching, creating, editing, and deleting memories.

  • Stats cards showing memory count and usage
  • Debounced search with date range filters and sort controls
  • Inline create, edit, and delete with toast feedback
  • Accessible from the chat sidebar navigation

Inline tool result cards

Every memory operation displays a styled result card directly in the chat conversation with mode-specific colors and icons.

  • Search results: blue cards with matching memories
  • Create confirmations: emerald cards
  • Update confirmations: amber cards
  • Delete confirmations: red cards
  • List results: violet cards with all stored memories

7 tRPC endpoints

Complete API coverage for memory operations through type-safe tRPC procedures.

  • Full CRUD: create, read, update, delete
  • Full-text search endpoint
  • Upsert for idempotent memory updates
  • List with organization scoping

Memory that actually works.

Not a proof of concept — a full-stack memory system ready for production.

Buy now
© 2026 hourzero. All rights reserved.