Blog
On AI memory, token efficiency, and running language models on your own hardware. RSS.
Memory is the token optimisation nobody is doing
Re-explaining context every session is usually the largest line in a token bill.
Memory · 2026-07-28How to make AI memory actually work
The failure mode of a memory system is not a full store. It is a store full of things that were true once.
Memory · 2026-07-28What belongs in AI memory, and what ruins it
Most memory systems fail on the write path, not the read path. They store the wrong things, and no error is ever raised.
Local-first · 2026-07-28Keeping AI memory on your own computer
Every product in this category is somebody else's server. There is another way, and a real trade-off attached to it.
MCP · 2026-07-28Connecting memory to any assistant with MCP
One protocol, every client. And a handshake with seven steps, six of which fail silently.
Memory · 2026-07-28Why your AI forgets everything between conversations
It is not a bug, and a bigger context window will not fix it.
Practice · 2026-07-28How to stop re-explaining your project every session
The preamble you type without thinking is the thing worth writing down.
Three threads
Memory — what belongs in an assistant's memory, how it goes stale, and why storing everything makes retrieval worse. Start with how to make AI memory actually work.
Tokens — where context spend goes, and why the biggest saving is usually not sending it at all. Start with memory as a token strategy.
Local-first — running memory on hardware you own, and the trade-off that comes with it. Start with keeping AI memory on your own computer.