Your AI Agent Forgets Everything. Here's the Fix by @oracledevs
From Kunal Kushwaha · · 2 min

The player loads from YouTube only when you click. It uses YouTube's privacy-enhanced mode.
If the video does not play, the owner may have turned off embedding. Watch it on YouTube instead.
Description
I've got like 10 emails in my inbox from companies building their own take on agent memory. It's a hot space right now.
What is happening? Your agent's context window resets between sessions, so whatever the user told it yesterday is just gone unless you save it somewhere. The obvious fix is resending the whole chat history every time, but that gets expensive quick and the model starts missing stuff that actually matters.
Met the Oracle team in San Francisco recently, they have a blog on how they went about this with their oracleagentmemory package.
👉 Read it here: https://fandf.co/3SDXjbR
A language model pulls the important facts out of each conversation, saves them per user, and on long chats the model gets a short context card instead of the entire transcript.
One number from their benchmark: 80 turn conversation, full history agent was at around 13,900 tokens per request. With memory, around 1,300. Same conversation, a tenth of the tokens. Sponsored by Oracle.