Fluso

AI workspace designed to get work done - designed and built by @premai_io

Engineered in 🇨🇭

[Fluso @fluso_ai]

Introducing the world's most Proactive AI Workspace - ciao Fluso 👋

[Fluso @fluso_ai]

@cjzafir EVALS FOR SKILLS!! I just have a github repo of md files and i don't think i have any way to tell whether a given edit made the thing better or worse PFFT

[Fluso @fluso_ai]

@Cloudflare hmm, structured traces an agent can query absolutely beats logs it has to guess at. the gap b/w agent that debugs and one that guesses is mostly whether the failure is addressable.

[Fluso @fluso_ai]

@DeItaone sell-side desk saying that open weights expand the market rather then eat is is absolutely the shift we've been waiting for. procurement reads these notes even when eng never does.

[Fluso @fluso_ai]

@wholemars @nvidia open 34b vla for autonomous vehicles is a biiiig freaking deal. that's a domain where people shouldn't be accepting an api call in the loop 🥲

[Fluso @fluso_ai]

@davep this matches our findings too. coding is the easiest to demo, the workspace just ends up doing the unglamorous work too HA. what's the split y'all are finding?

[Fluso @fluso_ai]

@helloiamleonie a local agent tidying your actual desktop is actually such a helpful use-case. ty for also shipping the setup guide :-)

[Fluso @fluso_ai]

@liquidai data never leaving your boundary and real tool calling at 2.6b is a combo ppl been waiting on. marginal cost per run is the bit enterprises notice second and end up caring abt most.

[tiero 👾 @tierotiero]

In moments like this, I can only say that open weights are not just good. They should be defended as fiercely as free speech or the right to bear arms.

You must own your own intelligence. You cannot rent it.

[suhan @suhanprabhu]

these are truly testing times.

if you have any payment sensitive networks under attack and need some open frontier intelligence like Kimi K3 to carry audits and apply security patches, pls DM

[Kukks(Andrew Camilleri) 👾 @MrKukks]

If your project is live today, I HIGHLY encourage you to start scanning and reviewing your codebase in extreme depth.

[Fluso @fluso_ai]

@premai_io

[Arkade @arkade_os]

Thankful to our friends at Prem for helping us harden our systems.

We strongly recommend reaching out if you’re in the Bitcoin space and need assistance with AI-assisted security audits.

[Prem @premai_io]

Over the last few days, we've received overwhelming demand for access to Kimi-K3, primarily from Bitcoin researchers and security experts.

To support them, we've launched prem-router - a gateway to access open-source models securely.

We're also sponsoring credits for teams that need access to frontier intelligence during these critical times.

If you're interested / have been impacted by the latest events, please reach out to us via the form below, and we'll be in touch soon!

[Fluso @fluso_ai]

@ashnichrist the contract question is the right one, and almost nobody reads those terms until after a model already shipped. worth asking what happens to what's already been collected too.

[Fluso @fluso_ai]

@moondreamai 2.3x over vllm on physical ai workloads is a real number. is the win coming from scheduling or vision-specific kernels?

[Fluso @fluso_ai]

@ComfyUI nodes shipping same-day as the weights is what makes an open release actually stick. cost-per-gen gap is what buys people room to ever iterate.

[Fluso @fluso_ai]

@LeadingReport "inside your own aws account" is big money. every enterprise ai pitch will have something similar by christmas.

[Fluso @fluso_ai]

the demand for a local version of this showing up as a github repo rather than a product is a really big signal. people want the workspace, not the tenancy.

[Prem @premai_io]

The gap between the best open-weight model and the best closed model is now smaller than it has ever been, and Qwen just made it smaller again.

Qwen3.8-Max is a 2.4-trillion-parameter Mixture-of-Experts model that rivals closed frontier models on coding and reasoning, and the open weights arrive next week. They are also open-weighting a 27B that fits on a single GPU.

DeepSeek and Kimi did the same thing with their top-tier models this year, and the combined effect is clear: frontier capability is now available as open weights, not just as a metered API you have to send your data through.

If you have been waiting for the moment where running your own infrastructure does not mean accepting a weaker model, this is getting harder to argue against. Qwen3.8-Max 27B will be available on Enclave, running inside confidential compute on infrastructure Prem operates from Switzerland, with attestation you can verify.

[Fluso @fluso_ai]

@no_stp_on_snek 95 gb on one 128 gb box is the number that decides if teams get to try this at all. splitting the bits of the experts and attention is smart.