𝗘𝘃𝗲𝗿𝘆𝗼𝗻𝗲 𝘀𝗮𝘆𝘀 𝘁𝗵𝗲𝗶𝗿 𝗮𝗴𝗲𝗻𝘁 𝗵𝗮𝘀 𝗺𝗲𝗺𝗼𝗿𝘆

𝗘𝘃𝗲𝗿𝘆𝗼𝗻𝗲 𝘀𝗮𝘆𝘀 𝘁𝗵𝗲𝗶𝗿 𝗮𝗴𝗲𝗻𝘁 𝗵𝗮𝘀 𝗺𝗲𝗺𝗼𝗿𝘆

People use the word memory for many different things.

When you say your agent has memory, you might mean one of these:

• Sending old chat history back into the context window.
• Querying a vector database for specific data chunks.
• Keeping a user profile that updates over time.
• Using a scratchpad to track tasks during a job.

Calling all these things memory is a mistake. They fail in different ways. They require different designs.

If you treat them as the same thing, you will build the wrong system.

What do you expect when you say your agent has memory?

Tell me what works for you and what fails.

Source: https://dev.to/jennapederson/everyone-says-their-agent-has-memory-26nj

Optional learning community: https://t.me/GyaanSetuAi