Hidde @piros_dev

Founding AI Engineer @ https://t.co/5eKWiqwvLF | prev: ML @Cambridge_Uni Joined November 2017

Tweets

2K
Followers

1K
Following

1K
Likes

4K

Hidde @piros_dev

2 days ago

Assistant-ui has a few for: composer (with submit, queue, being able to @ / in the composer etc) assistant-ui.com/docs/primitives which build on top of other radix-ui primitives under the hood I think Fluid functionalism or AI elements are the same. Everyone is re-inventing the same logic under the hood, while I would prefer to have an headless library that can handle this stuff, so people can focus on the styling part (while still being able to decide themselves what happens onSubmit) I fully understand that base-ui would probably want to focus on the lower-level primitives primarily, but I can see a world how this could set it apart from things like Radix. Like how the Drawer extends Dialog, or how people combine Dialog+auto complete for cmdk menus. (like an attachment could be similar to the Avatar component, where you have control over how you style the loading state etc) fluidfunctionalism.com/docs/ask-user-… elements.ai-sdk.dev/components/att…

1 0 0 36 0

View Details

Hidde @piros_dev

2 days ago

epic! yeah I don't think there are any good ai-primitive libraries rn. and ofc they can be made using other primitives, but then it feels like everyone is reinventing the same wheel. I don't know if that is in scope for base-ui, but would love some ai-primitives where the user would only need to decide on how they want things to be styled etc. Assistant ui and vercel ai primitives are kinda clunky and forgotten (unlike base-ui). And I think it would be good for a headless package like base-ui to have proper accessibility decisions baked in for common components like these x.com/piros_dev/stat…

Hidde @piros_dev

4 days ago

@elsigh @vercel easy way to set up storage (history, mayb memory could be cool like mastra). MCP, guardrails. branching etc. I love how easy it seems to set up the harness with eve, but actually deploying in in frontend production still seems to require like a lot of work (unless you deploy as

0 0 0 84 0

1 0 0 43 0

View Details

Hidde @piros_dev

2 days ago

@colmtuite @base_ui This looks epic! is this already possible and works for arbitrary triggers? ($, @, / etc), and how is this structured under the hood? like if you would send this message to an agent what would it see? Als need to be able to render these tags and skills in messages ofc.

1 0 0 33 0

View Details

Hidde @piros_dev

3 days ago

@fuma_nama Could be huge for companies that release product changelogs as well (instead of just package changelogs for oss)

1 0 1 229 0

View Details

Hidde @piros_dev

3 days ago

@devongovett need the first one in base ui @colmtuite 👀

0 0 0 219 0

View Details

Hidde @piros_dev

4 days ago

easy way to set up storage (history, mayb memory could be cool like mastra). MCP, guardrails. branching etc. I love how easy it seems to set up the harness with eve, but actually deploying in in frontend production still seems to require like a lot of work (unless you deploy as a slack agent). AI Elements feels like a forgotten child (and alternatives like assistant-ui are also not great) and could be integrated very nicely. This stuff is probably out of scope for eve, but would love to see a more opinioted headless components library (like base-ui), but for composer with / commands @ tagging, message queuing or steering. know you could set this up all manually, because you have the tool call type safety, but this is why it kinda felt more as it was built for "internal agents" or background agents, that the customer would not necessarily directly interface with. Amazing work regardless for the first release, and looking forward to where this goes!

0 0 0 84 0

View Details

Hidde @piros_dev

4 days ago

@FredKSchott @flueai Really appreciate the transparency here! Would love to hear what you think about mastra

0 0 1 113 0

View Details

Daniel A. Saedi (DataManDan) @TheRealDanSaedi

a week ago

Every sales and marketing tool now promises AI will improve your results but they don't tell you how it will cost thousands of dollars and hours of time to onboard We built a system with @OpenAI for the part everyone hides: onboarding And today we're giving it away for free

24 35 178 36K 60

View Details

Hidde @piros_dev

2 weeks ago

@roshunpatel this disclaimer will become the biggest flex in a year or two 👀

0 0 2 79 0

View Details

Daniel A. Saedi (DataManDan) @TheRealDanSaedi

2 weeks ago

AI can now read your customers' minds. We raised a $20M Series A lead by 8VC & Lingotto to build this. Introducing Minerva, built in collaboration with OpenAI:

562 386 2K 1.3M 2K

View Details

Hidde @piros_dev

3 weeks ago

@VictorTaelin thats opus for ya

0 0 0 54 0

View Details

Hidde @piros_dev

3 weeks ago

@Osamadev_ @thsottiaux 5.6

0 0 1 13 0

View Details

Hidde @piros_dev

2 months ago

@RCAgator @joshscorcher now replace the blue people with babies or elderly

1 0 0 193 0

View Details

Hidde @piros_dev

2 months ago

@ericzakariasson wish it could do this locally tbh.

0 0 0 65 0

View Details

Hidde @piros_dev

2 months ago

idk about this new Claude merch tbh

1 0 1 72 0

View Details

Hidde @piros_dev

2 months ago

@bridgemindai @sama still cheaper (but only a bit) if you take into account how much tokens each models use. new opus 4.7 tokenizer doesn't help. and 5.5 is very good on medium as well, barely use xhigh nowadays.