It has access to n a huge amount of personal notes and previous journal entries. The prompts make it "introspect" together with me and talk more deeply about my day and a wide range of topics. Then when the day is finished, all conversations are summarized into short paragraphs, along with the topics discussed, and some counter topics as well as counter arguments are added to the ongoing chat summary history. The next day, then these counter topics are in the mix of the entire history the assistant has access to (as vectors), and conversations start to steer away from the "bubble" which I usually journal and think in.
I also added some automation (morning and evening schedule) and randomness (5% chance of challenging my point as much as possible, 30% chance of telling me that it doesn't want to discuss the topic I picked to discuss, and that I should pick another one). I'm building this locally with Langchain, Redis & n8n and the OpenAI API and learning a lot through it. I have rewritten it completely twice now, and am slowly starting to see limitations and issues which I did not see previously.
We help people and companies to launch their projects :)