[hadlock]

The big draw of open claw is the memory architecture. Because you effectively start from scratch every time you open a new claude chat. Open Claw on the other hand, it compacts regularly, but also generates daily digests, and uses vector search, and then uses thoughtful memory retrieval techniques to add relevant context to your queries. Recent things get weighted more heavily, but full text search of all chats is still possible, and this is all managed automatically. Plus it uses markdown so the barrier to entry for manually auditing/modifying memories etc is very very low. If you say "can you check if my solar panel for my power generator arrive yet?" it is going to probably know what I'm talking about and go check my email for delivery notifications, based on conversations I've had with it about buying, ordering the solar panel etc. Claude is just going to ask clarifying questions since it has no idea what I am referencing.

[johnwheeler]

So it sounds like you get extra memory at the expense of having to compact more because, of course all those things are going to take up context. But since you’re not interacting with it in some kind of turn based fashion it makes it worth it— the lack of context doesn’t matter. Is that correct?

[hadlock]

Yeah it's basically just a smart compaction and retrieval algorithm, blended with vector search of uncompacted memories. The algorithm is open source, but the technology behind securing the agent against 1-shot prompt injection will not be.

[noelsusman]

I can just look at my front porch to see if there's a solar panel there, or failing that I can click a single button on my phone and search "solar" on my gmail and find out where my solar panel is. Having an agent do that for me saves me like... 5 seconds?

[hadlock]

Sure. I am (literally) currently feeding a newborn, my house is a diaster zone, and it's raining. DHL just changed my delivery date from today, to the 19th so maybe it will arrive today, maybe it won't. I haven't slept more than 4 hours in 3 days so getting an answer via voice memo seems pretty nice right now.