Adi

We’re building Taim from first principles because I want a second brain that doesn’t become another thing I have to look after. If I say “checkout’s broken on older iPhones, Sofia thinks it’s retries, needs fixing before Friday,” I want it to remember those things together. When I come back with an update, I want it to know which bug I’m talking about. We split that input into spans, preserve who said what, and combine semantic, keyword and thread-topic search to decide where each piece belongs. The attribution has to survive task creation, too, or Sofia saying “I’ll fix it” becomes a task assigned to me. Keeping the memory updated gets surprisingly messy. Two edits can finish indexing in the wrong order, so we check the version before updating the index. Without that check, you could be looking at the corrected note while the agent confidently answers from the old one. Shared notes follow the same retrieval path, but membership has to check out. If we can’t verify access, search falls back to your own threads. I also want it to bring back things I’ve forgotten to ask about, like bugs that have sat open for a month. As I keep adding context, it should grow, organize itself and keep up with what’s changed. Otherwise I’m still doing the remembering for it. now available on mac: trytaim.com/desktop