Seni

Seni

A cognitive architecture
#100242151 follower0 following
#1,709Last month
20 KP
>10,000All time
21 KP

About

solo builder, working on Ash, a cognitive operating system that runs entirely on your own computer instead of the cloud. Started because I was drowning in email, tasks, and context-switching myself and got tired of cloud AI tools that don't remember anything past one conversation and want your data on their servers. Turned into building the real thing: local LLM, persistent memory across days/weeks, autonomous task follow-through, browser automation, the whole stack — solo, from scratch. Systems thinker, which mostly means I care more about does it actually work reliably than does it look go

Badges

Tastemaker
Tastemaker
Gone streaking
Gone streaking
Gone streaking 5
Gone streaking 5

Maker History

Forums

•

24d ago

Ash - The AI that actually does the work, not just tells you how

Not a general chatbot built for people whose time gets eaten by the repetitive layer under their real work. **Solo consultant:** Inbox and calendar admin eat billable hours. Ash handles the admin layer. **Agency owner:** Juggling calendars/clients. Ash catches conflicts as they happen. **Founder wearing every hat:** Ash takes inbox, scheduling, research off the stack. **Realtor/field operator:** Client follow-through doesn't stop on the road. Ash keeps admin moving without a laptop.
•

1mo ago

Is the hardware requirement the real barrier to local AI adoption, not the software?

Been deep in the weeds this week on what it actually takes to run a serious local AI setup not a toy chatbot, something that actually does real automated work. The honest number ended up being 24GB+ VRAM minimum, 48GB+ recommended for a smooth experience. That's not "install an app," that's "own real hardware."

Which makes me wonder if local-first AI has a much smaller realistic audience than the privacy pitch suggests plenty of people want their data to stay local, but way fewer actually have (or want to buy) a machine that can run it well.

I've been building Ash, a local AI agent, partly to test this myself but genuinely curious from others here: if you've looked at local AI tools, was the hardware bar the thing that actually stopped you, or was it something else (setup complexity, model quality, just habit)?

•

1mo ago

Anyone else noticed the same LLM give a completely different answer to the identical prompt, twice?

Ran into something this week that's been bugging me: gave a local model the exact same prompt, replayed it twice with nothing else different, and got a correct answer once and a wrong one the second time. Not a hard edge case either a pretty basic task it clearly CAN do, it just didn't do it reliably both times.

Made me rethink something: a lot of what gets blamed on "the model isn't good enough" might actually be a consistency problem, not a capability ceiling. Those are very different things to fix capability needs a bigger/better model, consistency might just need lower temperature, better prompting, or accepting you need a verification step regardless of model size.

Curious if others building with LLMs (local or cloud, doesn't matter) have run into this do you design around inconsistency (retries, verification passes, structured output constraints), or has this mostly not been an issue for what you're building?

View more