AI memory vs. chat history
Decide when you need the original conversation and when a short memory record is enough, with a checklist for checking exact claims.
Understand AI memory 2 min read
What is AI memory?
AI memory is information saved for use beyond the current exchange. A system may keep whole conversations, selected facts, or summaries. What it stores and what an assistant actually reads are separate questions.
A saved conversation is only the starting point
Suppose you tell an assistant that a launch has moved to November. That sentence might remain in a transcript. A memory system could also extract the new date as a fact. Later, a search step has to find the relevant record and put it in front of the assistant.
A fluent answer does not prove that any of those steps happened. The assistant might be using the current chat, an old note, or a guess. Ask which source supports a detail when the distinction matters.
Keep facts with their boundaries
“Use a short, direct style for the launch email” applies to one piece of work. It does not necessarily mean you want every future answer to be short. A useful memory keeps that scope, along with the date and any source that lets you check it.
- Save the project or task the fact belongs to.
- Keep decisions separate from ideas that were only discussed.
- Mark facts that have changed instead of silently treating both versions as current.
TAKE THIS WITH YOU
A memory worth checking
An illustrative template. Replace the details with material you are allowed to share.
Project: November launch Decision: Move the announcement to 12 November. Status: Confirmed by the project owner. Source: Launch planning chat, 16 September. Replaces: The earlier October target. Still open: Whether the beta starts before the announcement.
USING THIS WITH DRAY
Keep the context. Check the source.
Dray processes captured conversations into memory records and supporting references. Review the returned context against the source; memory can be incomplete or wrong.
Dray is in beta. Available sources and connections depend on your setup. Read the data notes before granting access.
Sources & next reads
Product details are grounded in the guides below. Example projects and prompts are illustrative, not customer results.