Long-term memory
In one sentence
Information an AI retains across multiple conversations, such as user preferences, past decisions, and completed actions.
Large language models do not remember previous conversations on their own. Long-term memory works by having software store key information and reload it into context during the next conversation.
Memories must be reviewable and deletable; outdated or incorrect memories will continue to distort future answers.
Example
An employee told the AI agent last week, 'Always format reports in Word.' When asked to write a report this week, it outputs a Word file directly.
Related terms
- Context window
- The maximum total amount of text an LLM can process at once, including prompts, reference materials, chat history, and its own generated responses.
- LLM Wiki
- One-page referenced briefs compiled by AI from notes, files, and meeting transcripts to answer 'What has been decided?' and 'Which version is the latest?'
- AI Agent
- An AI that can read files, call external tools, and complete multi-step tasks on its own, unlike a standard chat AI that only replies with text.
GEO Guide
How Domestic Brands Enter the Long-Term Memory of AI Models
