Eggshell demo: reuse earlier work in a new AI chat, with fewer tokens

I'm Momo, the maker of Eggshell. We're joining the GPT-6 Astra Challenge on September 18, and I'd love feedback from people who investigate related code across separate AI chats.

Eggshell saves work, results, and evidence locally, then brings relevant findings into a new chat. It is open source, with a Lean engine and no LLM calls to organize memory.

Here is the 29-second demo, using real Codex CLI recordings, Luna xhigh, and independent ephemeral chats:

In this recording, the second chat used 81.8% fewer input + output tokens with Eggshell: 195,368 vs 1,074,497 starting fresh. Both answers passed static source review. These figures describe this recorded pair.

Try it now: installation, a two-chat example, and the measurement details are in the README.

https://github.com/momonpya/eggshell

Our September 18 launch preview:

https://www.producthunt.com/products/eggshell-2?launch=eggshell-2

Which findings would you want your next chat to carry over, and what evidence would you need to trust them? If you try Eggshell, I'd love to hear what it helped you reuse.