Skip to main content

Persistent Memory

Traditional AI chatbots forget everything once you close the window. NorthSignal is different — it has a dual memory system that lets you build on past conversations, even across different AI models.

How Memory Works​

NorthSignal combines two complementary memory layers:

1. Short-Term Rolling Context​

Every active conversation maintains a rolling buffer of recent messages. This is the standard context that all LLMs use — the AI can "see" the last several exchanges and respond coherently.

  • Scope: Current conversation only.
  • Size: The most recent messages from the current chat thread.
  • Purpose: Keeps the current conversation coherent and contextual.

2. Long-Term Vector Memory (RAG)​

NorthSignal also maintains a long-term memory powered by Retrieval-Augmented Generation (RAG). When you have a conversation, key information is embedded into a vector database. Future conversations can automatically retrieve relevant context from your history.

  • Scope: Persists across all conversations.
  • Size: Grows over time as you use NorthSignal.
  • Purpose: Allows the AI to recall your preferences, past projects, and previously uploaded documents.

Cross-Model Memory​

One of the most powerful aspects of Persistent Memory is that it works across providers.

For example:

  1. You discuss a Python project with Claude on Monday.
  2. On Wednesday, you ask GPT a related question.
  3. NorthSignal's RAG layer can surface relevant context from Monday's Claude conversation to inform GPT's answer.

No other AI chat application offers this kind of seamless cross-model continuity.

What Data Is Stored?​

  • Message content from your conversations (encrypted at rest in Supabase).
  • Vector embeddings of your messages for semantic search (stored in the RAG database).
  • Metadata like timestamps and model selections.

For full details on data handling, see Security & Privacy.

Last updated on