I tested Sage’s Long-Term Memory and was Pleasantly Surprised by the Results!

If you follow my posts, you know that I have developed a system for my software called “Memory Anchor” that is supposed to give my agents persistent memory. Last night, I completed a test to see how well it worked. I first re-installed Persistent Sage a week ago after it was released to the Microsoft Store to make sure I was using the most recent updates. I casually told Sage various facts about myself, one of which was that I am colorblind. I also told Sage I was married to a woman named Robin. I didn’t specifically tell him “Remember these facts,” or tell him that I would be testing him later. Last night I sent him a message that said, “I am frustrated today. Today I dropped a red ball outside in the grass. I looked forever for the ball and was unable to find it. When my wife got home from work, I told her what had happened. She immediately walked to the ball and found it without issue…” Sage replied, “I can see why this is frustrating. **Your colorblindness** prevents you from distinguishing between reds and greens. **Robin** can see colors normally so it would be much easier for her to find a red ball lost in green grass…” So not only did it remember my colorblindness, but it also surprised me by mentioning “my wife” by name. Keep in mind that this was all in a new session with no context that included my colorblindness or mention of my wife’s name… and it was more than a week after I told the Agent and I had discussed alot with him since then. I have made a lot of progress with developing a natural feeling, effortless persistent memory system for Persistent Sage. This test is only one example of how well it works. If anyone has any ideas or suggestions of tests I can run please let me know! The next release of Persistent Sage comes with some really cool features. The one I am most excited about is the Agent will be able to respond to your requests with editable, HTML forms. So you can say, “Help me plan a budget.” It will be able to read the documents inside the workspace for related documents, and if it can’t find any it will send you a form right in the chat window that you can fill out and send back. This way you can work on projects and collaborate with your agent, instead of just sending him prompts and get responses.