Following the trail with the Omnesis Graph
How the Omnesis Graph allows efficient retrieval of adjacent context to answer user queries
BUILDING OMNESIS
I’ll try to share updates and interesting stories of how I went about building Omnesis.
How the Omnesis Graph allows efficient retrieval of adjacent context to answer user queries
From source-declared aliases to the people graph, automatic merge heuristics and an experimental Brain that can investigate whether two identities belong to the same person.
The Omnesis Chrome Extension allows capturing the web pages I see. I evolved it also help me find URLs I care about
A walkthrough of the inner workings of Omnesis' search engine and a short story of recent performance improvements
Omnesis transcribes voice notes sent on WhatsApp/iMessage as well as notes you dictate on your phone and watch. I’ve been working to leverage the context that Omnesis has about the author of the voice message to improve transcription.
Omnesis has a background agent that processes years of emails, messages, calendar events and other personal data. Bootstrapping 3.5 weeks of my own history required 640 agent runs and cost about $20. Today I got that down to $6—mostly by realizing I'd structured my prompts badly for provider caching.