I spoke at Hermes AI Agent Night: The AI Operating System at SGInnovate in Singapore, with thanks to Lionel Sim from The AI Capitol for organising and An Ting Lim from SGInnovate for hosting the community.
Great to be able to hear from Tommy Eastman from Nous Research and share the stage with Ashish Verma, who demoed his open-sourced 8mem memory layer for multi-agent set ups.
I shared how I use Obsidian as a company wiki and demoed it for Sales and GTM professionals: context, meeting notes and follow-ups all in one local database. I also walked through Karpathy's LLM Wiki, modified over the weekend with Dr Vivian Balakrishnan's OKF idea to save tokens as a rulebook for my agents' second brain.
Over the last few weeks, the same questions kept coming up: what do I use as my second brain and how do I handle data privacy and security. My Obsidian wiki is only reachable by devices I let onto my secure Tailscale network. For confidential data, I run local models to turn messy inputs (emails, voice notes and so on) into a structured Karpathy-style wiki, building it from scratch on the machine with no third-party LLM or service looking at the data.
We keep seeing models go off the rails in the news, acting without a human in the loop when instructions are not bounded by clear safeguards. Full control over how data is used is not a nice-to-have, it is a necessity and it matters most for enterprises, especially those in regulated industries.
Good to see builders in the community already running Obsidian fully local and starting to sync devices with the free, community-built LiveSync plugin.
My full slide deck is here: obsidian.procss.online