The memory outlives the tool.
Memkith is a shared memory layer for AI coding agents. It preserves decisions, constraints, failed attempts, and review outcomes, then serves them back to whichever agent is about to touch the code they concern.
Every coding agent ships some form of recall, and all of it dies at the tool boundary. Switch from one agent to another and the accumulated context does not come with you. Memkith holds one team history that every connected agent reads and writes, so the agents stay interchangeable and the memory stays put.
Two things follow from that. Memory is addressed by code rather than by conversation, so retrieval is keyed to files, functions, and lines. And memory inherits code visibility, so it follows the same lifecycle and the same access rules as the work itself.