Post

So after running this a while, I can confidently say that Modal is def. not going to break the bank. I am using it all the time (from mobile, laptop, claude code cloud, claude dot ai chat) and I'm averaging about $0.10/day (plus about $0.02/day on Qwen3 embeddings). fast, cheap, globally reachable

i haven't gotten into stateful agents, what are the biggest advantages you see using this over just Claude Code alone?

It means that my ADHD stream-of consciousness which naturally spreads across many projects (some of whom touch on each other) and surfaces (idea comes to me when on my phone in a parking lot...), plus random "ah need to remember to get milk!", can all get baked into "me and Claude"

every conversation flows from one to the next, and I have hooks that tells Claude to check for relevant other stuff - which is sometimes annoying "this would be great for your talk next week!" (no, it won't, I've got that under control thx), but often helpful context so I don't keep repeating myself

and when I need to have a very focused, specialized Claude, who is On Topic? My stateful yaucca-enabled Claude can spin off a sub-agent with a specialized prompt for just that task, and none of my stateful-agent memories and other stuff cluttering up its context. Best of all worlds!

how similar is it to using Chatbot Claude with memory enabled? same thing, but for the agent, or better in some way?

a) important, named, updateable memory blocks, *always in context* b) most recent N conv. messages, so I literally can switch from any conversation or surface and ask "what was I just saying" and *without a tool call*, it's literally already above my message in the context c) archive memory MCP

it's basically the MemGPT architecture ( arxiv.org/abs/2310.08560 ), which the Letta team (who wrote it) went and extended and turned into a whole thing.

what's the advantage of Yaucca over letta?