A github awesome-list. Curated this because all other resources I found mix between AI and SI. Suggestions are welcome, thanks!

I'll be around, happy to bring you up to speed with your first agent or answer anything else

Thanks! in the works, you can partly go around this now by piping out and back in but we'd like to make it smarter - llm wiki and graph.

Adding to Shaked’s point: persistent memory can become stale and deciding what to retain and trust is a security trade-off. The current stateless approach, combined with the code sandbox, can rebuild context quickly. The idea is to add smarter memory without losing that property.

I'll be around, happy to answer anything.

On a subscription Claude Code already defaults to 1h TTL and on an API key you opt in with ENABLE_PROMPT_CACHING_1H=1.

Was the ~22% claimed in the repo is measured against the 5m default, or with the 1h flag on?

Looks interesting. How does the behavioral analysis work?

+1, I don't understand what it does.

What's the advantage vs the providers' native similar features?

First, it's terminal first, meaning you can use your existing subscriptions and take advantage of any and all the different provider CLI's.

Second, it's a persistent workspace so you aren't locked in to whatever features the provider's build. Instead you can really drive your own workflow as you see fit.