ChatGPT memory helps.
Local MD files help.

But neither travels cleanly across everything I use, and packing too much into MD files eats context and tokens.

With Empirical, I keep my AGENTS.md lean and let Codex pull context dynamically when it actually needs it.

I can open ChatGPT on my phone, connected to Empirical, and it pulls the same memory context and writing tone I use in Codex or any other connected AI tool.
That means:

• less repeated setup
• cleaner, cheaper prompts
• more consistent output across sessions

This is just the tip of the iceberg.

I wrote up a Codex example here: