VecFS is short for Vector File System or Spec - it's both.
Imagine being able to give your AI agent long-term memory by storing context locally in a manner most efficent and available for Agents. This could allow agents to learn and improve over time because they might recall past mistakes.
VecFS should provide a local file that is simple to operate, back-up and maintain and allow everyday users to extend the learning and memory of their AI agents in a simple way, without the complexity of running vector databases with extensive storage.
Model Context Protocol allows for the integration of context into AI Agent tools. Vector Databases are typically too large to run on a laptop but you don't need all the dimensions in a vector DB all the time.
A natural compression can be achieved with right-sized thinking in the mix.
That hypothesis is that we can help agents be more useful and more efficient with long-term recall AND that a vector information system does not need all the dimensions for all of the objects all of the time.
The inspiration for the natural data compression is Look Up Tables (LUTs) in older image file formats for systems that could not express all of the colours all of the time because there were too many bits per pixel.
You don't need to store zeros.
Allow the vector space to be dynamic and declarative and for each vector to indicate the dimensions, where it has non-zero values, leaving others as implicitly zero.