Marcin Uchacz

building one context layer for all apps, LLMs and agents @

[Marcin Uchacz @marcin_u2]

@NotionHQ imo it's because you don't need just a strong model or agent to get to Level 3/4, but the whole infra (context and memory, permission guardrails etc)

otherwise, ppl are just scared of allowing agents to execute core workflows.

[Marcin Uchacz @marcin_u2]

@ky__zo @OpenAI @github lfg (as always)

[Marcin Uchacz @marcin_u2]

@michuk you can just plug your tools and LLMs to Unabyss, and it becomes a shared memory layer across all of them

faster than vibecoding, cheaper in a long term + likely much better when it comes to retrieval precision

without a line of code, in fact

[Marcin Uchacz @marcin_u2]

@ky__zo @michuk thx, indeed!

so Unabyss solves this problem without any line of code - you just plug in your tools and LLMs, and it becomes a shared memory layer across all of them

faster than vibecoding, cheaper in a long term + likely much better when it comes to retrieval precision

[Marcin Uchacz @marcin_u2]

@buildinpublic Connecting our context infrastructure to a few open source memory systems (Gbrain and LLM Wiki)

[Marcin Uchacz @marcin_u2]

@Joi2James https://t.co/xkjNkxVnrY - we just need 100 more 😂

[Marcin Uchacz @marcin_u2]

@markproduct Putting on the mask that you know exactly where you're going. In front of investors, your team, and even yourself. And that mask is part of the job as a founder :)

[Marcin Uchacz @marcin_u2]

@vitaliidodonov congrats! we've actually done a similar launch before (also #1 of the Day, #2 of the Week).

we'll be creating a Product Hunt tutorial soon. happy to share best practices both ways!

PDF

[Marcin Uchacz @marcin_u2]

@apnmrev I think it's a problem of GTM knowledge being inaccessible for most. You have thousands of coding / vibe-coding tutorials, while GTM folks rarely share their tactics. Yet, there are a few of them doing it in a proper and transparent way (look at Growth Engine, Revpack and more)

[Marcin Uchacz @marcin_u2]

to make AI better: models or memory?

we're past the point where model intelligence is the bottleneck.

"Are We Ready For An Agent-Native Memory System?" shows the memory layer matters more than which model you plug in

and we see it a lot at unabyss.

[Marcin Uchacz @marcin_u2]

@heyblake https://t.co/xkjNkxVnrY

[Marcin Uchacz @marcin_u2]

@sridharfyi https://t.co/xkjNkxVnrY

[Marcin Uchacz @marcin_u2]

@CryptoMavka The only thing with Wiki (which I love btw) is that it doesn’t do well with large datasets, eg from email, slack or crm, plus, lacks native connectors to the above.

[Marcin Uchacz @marcin_u2]

@ColinGardiner emailed!

[Marcin Uchacz @marcin_u2]

@suzatweet splitwise openclaw for me and my cofounders to help us order more pizza and know who owes whom

[Marcin Uchacz @marcin_u2]

@sridharfyi https://t.co/xkjNkxVnrY

[Marcin Uchacz @marcin_u2]

@sridharfyi https://t.co/xkjNkxVnrY -> a self-updating context silo for all your apps and LLMs

[Marcin Uchacz @marcin_u2]

@16vchq agents will change, context won't - separate it.

[Marcin Uchacz @marcin_u2]

European founders are screwed

worse ecosystem, less capital, non-native English speakers + they grew up with a work-life-balance mindset

competing with the Bay Area is almost impossible

fortunately, Eastern Europe is built differently 🔥

[Marcin Uchacz @marcin_u2]

@kocer_eth the one thing it misses though it's keeping context up-to-date and other maintenance functions, so you can use unabyss for free to take care of that (we built custom connector)