Private Super Intelligence
Earth
Joined May 2023
- Tweets1.5K
- Following78
- Followers4.3K
- Likes471
Last year’s Plan ₿ week in Lugano was a huge success and this friday, we're set to welcome you to our own headquarters for an exclusive unveiling of what we have been building!
If you caught our invitation last week, we hope you’ll join us. After the unveiling, you can try the products and talk with our researchers and engineers about what you’d like to do with local and confidential AI.
We’ll keep the conversations going over aperitivo and live music. Looking forward to seeing you in Lugano!
23 October · 16:30–20:30 CEST
Prem HQ, Lugano
Request your place today, 19 October · Approval required. Register through the link below!
DeepSeek just released one of its strongest models for coding and agents, in the name of a “minor update.” 🫣
Here’s what got us most excited about DeepSeek-V4.1-Flash-1017:
- A 1M-token context window, with code and weights released under MIT.
- 77.0 on DeepSWE, ahead of Opus 5 at 74.0 and GPT-5.6 Sol at 73.0.
- 60.5 on Automation-Bench, up from 54.8 in the previous version.
- 42.3 on Terminal-Bench 4.0, up from 31.2.
Those are the published results. The team also reports faster task completion and shorter outputs, while keeping its API pricing unchanged.
Love to see open-source AI get this good, and we’re working to bring this release to Prem very soon. We’ll share access details once it’s ready.
Congrats to the DeepSeek team!
We’re opening the doors to Prem HQ in Lugano, and giving a small group a first look at what comes next.
Prem Unveiled brings together our work on local AI and confidential compute, along with our latest product. We want you to have a say in where your AI runs and how your data is handled, and this is a chance to see the work behind that commitment.
Join us for the live unveiling, try the products and speak with the researchers and engineers building them. Guests will also receive early access to upcoming releases, with aperitivo and live music to close the evening.
23 October · 16:30–20:30 CEST · Lugano
Places are limited and attendance is subject to approval.
We look forward to welcoming you to our home in Lugano and sharing this next step for Prem with you in person.
Prem has successfully completed its SOC 2 Type II audit. We’re proud to share this with everyone building with us and considering Prem for their team.
When you bring AI into your work, you’re also choosing a company to trust with part of that work. You need to understand how it manages security and whether its safeguards hold up in everyday operation.
SOC 2 Type II provides an independent examination of that. An external auditor reviews the design of the controls included in the audit and tests how effectively they operated over a defined period. Those controls are the processes and technical safeguards used to manage risks to systems and data.
The resulting report documents what was covered, how it was tested and what the auditor concluded. Your security and procurement teams can use that evidence when evaluating Prem against your organisation’s requirements.
We build private AI because people should be able to use it with control over their data. This audit gives customers additional evidence to examine as they decide how Prem fits into their work.
Thank you to the Prem team for the care and work that went into completing this audit.
A woman in Florida was arrested after allegedly threatening a sheriff’s office in a conversation with Claude. She reportedly used the chatbot as a diary.
According to the report, her messages were flagged, reviewed by a person and passed to police.
A chat can start with a small question and turn into a conversation about your health, your relationship or anything personal for that matter. And you may not realise that someone at the company could also read what you’re sharing.
Over the past few months, multiple reports and investigations have documented AI companies storing personal information, having people review conversations and using some of that data to train models.
These details aren’t always explained clearly to the people using the app. Someone opening a chat to talk about their day shouldn’t have to search through a privacy policy to find out who can read their messages or how they might be used. Companies need to explain that in plain language, before people start sharing personal information.
In Fluso, you can turn on Confidential Mode to protect your information while the AI processes it. The model runs inside a hardware-protected environment designed to prevent the people operating the servers from reading the information inside it. Prem’s confidential inference service doesn’t retain the request contents after processing or use them to train models.
If you’re building AI features that use private code, client documents or any sensitive data, you need to know where that information can be read while the model works with it.
We don’t use your API prompts or responses to train models.
With Prem’s Enclave API, the TypeScript SDK or local Confidential Proxy encrypts your request before sending it to our gateway. The gateway uses metadata, such as model choice and request size, to handle the call while the contents remain encrypted.
On supported confidential routes, the request is decrypted inside a hardware-isolated enclave for inference. The hardware is designed to restrict the host’s access to the data while the model processes it.
This limits where the sensitive content is readable, including keeping it encrypted as it passes through our gateway.
Our quickstart walks you through your first encrypted request. You can also explore how the data is processed, what’s protected and the current security limits.
Get started: docs.prem.io/quickstart
Today, we’re excited to launch Premonition-Exp!
We wanted to bring Jev-like decisions to audio workflows. A request can arrive as a voice note, with the details needed to evaluate it in a document or screenshot. We wanted a model that could work with those inputs together.
So we added audio to DiffusionGemma. Premonition-Exp accepts text, images, video and audio, including mixed inputs in the same request, and returns structured decisions.
For example, you can give it a spoken purchase request alongside an approval policy and ask whether it meets the requirements or needs review.
We trained a 4.3M-parameter projector to connect an audio encoder to DiffusionGemma, keeping both the encoder and backbone frozen.
It’s experimental, and we’re releasing the weights and model code for others to build on.
Watch the demo and find the weights and setup instructions on our Hugging Face model card: huggingface.co/prem-research…
Jev-compatible decisions are going beyond text.
We’re bringing DiffusionGemma to Prem’s API for questions about text, images and short video, with zero data retention.
See how it works and try your first API call.
We can finally share this! We’re bringing Bitcoin Red Team together with @LuganoPlanB on 22 October, right here in our home city.
Super excited & looking forward to getting Bitcoin developers and security researchers in the same room to talk through how AI can help with security reviews, what still needs expert judgment, and what we can learn from recent incidents.
Joining us are @Rob1Ham , @danielabrozzoni, @KLoaec, @MrKukks and @giacomozucco , alongside Federico N. and @suhanprabhu from Prem AI.
If you build, maintain or review Bitcoin software, we’d love to have you there. Bring your questions and experiences.
22 October · 15:30–18:30 CEST
@pow_space, Lugano · Sessions in English
Request an invitation through the post below. Space is limited, and the hosts will review requests and confirm places.
What scares Bitcoin security experts right now? Not FUD actual threats they lose sleep over.
Listen to the experts on the front line @KLoaec, @mrkukks, @Rob1Ham, @danielabrozzoni, @giacomozucco, @suhanprabhu and others together on Oct 22, 15:30 – 18:30 at @pow_space.
What threat keeps you up?
Powered by @PremAI_io
Learn more → planbweek.com/events/red-tea…
Prem retweeted
Who controls your AI infrastructure?
Your data, your models and where they run are becoming increasingly important questions.
@suhanprabhu works on exactly that at @premai_io, helping organisations build private and verifiable AI systems.
Join him in Lugano to explore what sovereign AI looks like in practice.
📍 Plan ₿ Forum | Oct 23–24
We’ve brought our abliterated DeepSeek model into Fluso's Confidential Mode.
Here’s what reduced refusals and encrypted inference mean for researchers, and how to try it with papers from your reading list.
Prem Confidential API's and ZDR API's have already processed 3.8 billion tokens across 124,000 requests and 11 models.
None of those 3.8 billion tokens is sitting on a server anywhere. Your prompts and completions aren't written to a log or a cache, and they're not used to train anything, so whatever you're working on stays yours.
If you've been wanting to try a strong model but didn't love the idea of your sensitive data living on someone else's server, ZDR is for you.
