Building AI-sovereign in Europe at https://xitter.cf/t.co/Y5tzLh08nF

Alicante, Spain
Joined January 2010
Oliver Montes retweeted
El Salvador deploys. We are so far ahead of other countries. And this is just the beginning.
MTS
@MTSliveAug 28
a16z's @jkhamehl says the US is being outpaced on AI adoption by countries you'd never expect: Bukele put Grok in schools for free while America fights over data centers "El Salvador, President Bukele has been super aggressive around utilizing AI. They implemented Grok in their schools for free. They're utilizing AI doctors." "City-state countries like Singapore or UAE, where it's easier just given the population to adopt technologies and then also make it subsidized or free. You see these different examples around the world where they're accelerating their AI development and adoption way faster than in the US." "In the US we've been seeing a lot of political tail-headwinds with data centers, and in some instances around surveillance, that are just falsehoods." "@elonmusk's building data centers in space. We might also see the influx of a lot of this data center supply chain build-out happen overseas because of the sentiment in the US." @a16z
5
18
2,158
Oliver Montes retweeted
Hi Oliver, 50 agents alone takes up a lot of RAM. e.g. a single Grok or Claude Code agent can up to 4GB of RAM We've kept the core Orca app really performant and built a resource manager to help you track agents's CPU & RAM usage xitter.cf/orca_build/status/2077…
If your laptop sounds like it’s about to take off, your coding agents might be the reason 😅 This Claude Code agent was casually using 4+ GB of RAM—and plenty of CPU. Orca shows you how much CPU and memory every active agent is using. Just pop open the 'Resource Manager' in the bottom-right corner 👇
1
2
118
Oliver Montes retweeted
We worked in close collaboration w/ @PyTorch & TorchAO teams to make offloading work with fancy quants 🔥 Consumer GPU users can now benefit from the goodness of modern quants like FP8, NVFP4, while keeping memory at bay 🤗 AND you don't have to give away latency! 📝 in ⬇️
3
7
44
2,852
Oliver Montes retweeted
Andrej Karpathy
@karpathyMar 31
New supply chain attack this time for npm axios, the most popular HTTP client library with 300M weekly downloads. Scanning my system I found a use imported from googleworkspace/cli from a few days ago when I was experimenting with gmail/gcal cli. The installed version (luckily) resolved to an unaffected 1.13.5, but the project dependency is not pinned, meaning that if I did this earlier today the code would have resolved to latest and I'd be pwned. It's possible to personally defend against these to some extent with local settings e.g. release-age constraints, or containers or etc, but I think ultimately the defaults of package management projects (pip, npm etc) have to change so that a single infection (usually luckily fairly temporary in nature due to security scanning) does not spread through users at random and at scale via unpinned dependencies. More comprehensive article: stepsecurity.io/blog/axios-c…
🚨 CRITICAL: Active supply chain attack on axios -- one of npm's most depended-on packages. The latest [email protected] now pulls in [email protected], a package that did not exist before today. This is a live compromise. This is textbook supply chain installer malware. axios has 100M+ weekly downloads. Every npm install pulling the latest version is potentially compromised right now. Socket AI analysis confirms this is malware. plain-crypto-js is an obfuscated dropper/loader that: • Deobfuscates embedded payloads and operational strings at runtime • Dynamically loads fs, os, and execSync to evade static analysis • Executes decoded shell commands • Stages and copies payload files into OS temp and Windows ProgramData directories • Deletes and renames artifacts post-execution to destroy forensic evidence If you use axios, pin your version immediately and audit your lockfiles. Do not upgrade.
559
1,091
217
10,423
1,634,337
Oliver Montes retweeted
Andrej Karpathy
@karpathyMar 24
Software horror: litellm PyPI supply chain attack. Simple `pip install litellm` was enough to exfiltrate SSH keys, AWS/GCP/Azure creds, Kubernetes configs, git credentials, env vars (all your API keys), shell history, crypto wallets, SSL private keys, CI/CD secrets, database passwords. LiteLLM itself has 97 million downloads per month which is already terrible, but much worse, the contagion spreads to any project that depends on litellm. For example, if you did `pip install dspy` (which depended on litellm>=1.64.0), you'd also be pwnd. Same for any other large project that depended on litellm. Afaict the poisoned version was up for only less than ~1 hour. The attack had a bug which led to its discovery - Callum McMahon was using an MCP plugin inside Cursor that pulled in litellm as a transitive dependency. When litellm 1.82.8 installed, their machine ran out of RAM and crashed. So if the attacker didn't vibe code this attack it could have been undetected for many days or weeks. Supply chain attacks like this are basically the scariest thing imaginable in modern software. Every time you install any depedency you could be pulling in a poisoned package anywhere deep inside its entire depedency tree. This is especially risky with large projects that might have lots and lots of dependencies. The credentials that do get stolen in each attack can then be used to take over more accounts and compromise more packages. Classical software engineering would have you believe that dependencies are good (we're building pyramids from bricks), but imo this has to be re-evaluated, and it's why I've been so growingly averse to them, preferring to use LLMs to "yoink" functionality when it's simple enough and possible.
LiteLLM HAS BEEN COMPROMISED, DO NOT UPDATE. We just discovered that LiteLLM pypi release 1.82.8. It has been compromised, it contains litellm_init.pth with base64 encoded instructions to send all the credentials it can find to remote server + self-replicate. link below
1,332
5,205
1,316
27,501
66,667,782
Oliver Montes retweeted
Fun fact: OpenAI handles 800 million users on ChatGPT with just one PostgreSQL primary and 50 read replicas 🤯 Today, OpenAI published an engineering blog explaining how they scaled their Postgres setup to support a massive 800 million users using a single primary and 50 multi-region replicas. They dive into details around their scaling approach, the PgBouncer proxy, cache locking, and cascading read replicas. It is genuinely neat and impressive. I just published a video on my YouTube channel where I dissect the blog and break down the nuances. Give it a watch - it is short and fun.
172
707
104
9,867
1,662,036
Oliver Montes retweeted
🟠 You can now connect Supabase to Levante in just a couple of clicks. Thanks to OAuth integration, setting up your backend is now: -> Fast -> Secure -> Zero manual configuration Less friction. More building. 💙 Levante is open source — contributions, feedback, and stars are welcome. --> Github repo: github.com/levante-hub/levan… --> web: levanteapp.com/
3
5
158
One of the few clients that support MCP-UI. With MCP-UI integration, servers can return interactive JavaScript components, like this game rendered securely in the client. Dynamic UI, zero friction.
Meet Levante: the open-source MCP desktop client built for privacy. Run AI locally. Connect any model. Your data stays on your machine.
1
1
72
Oliver Montes retweeted
Launching this Thursday, Nov 6, 2025 → Levante v1.3 open beta A local MCP client that makes MCPs practical for developers. Local-first. Multi-provider (OpenRouter, Vercel Gateway, Ollama, Claude, ChatGPT, Groq, xAI, Google AI). Want to collaborate? Get early access → levanteapp.com Repo → github.com/levante-hub/levan…
2
1
74
Meet Levante: the open-source MCP desktop client built for privacy. Run AI locally. Connect any model. Your data stays on your machine.
4
1
1
10
465
Oliver Montes retweeted
Guillermo Rauch
@rauchg24 Nov 2025
LALIGA is trending again, so it's worth giving an update. We previously wrote about how this soccer league in Spain was granted broad internet censorship powers[1]. 1️⃣ Vercel's customers have been unaffected We've taken drastic measures to ensure the uptime of our customers. While we rejected LALIGA's broad approach, our goal at Vercel is to protect and maximize our customers' and developers' freedoms within the limits of the law. We gave them a dedicated email inbox and an incident response automation. We have instructed our on-call SRE to expedite the review of these reports, because they can result in the loss of availability of entire sections of other, law-abiding customers. This is what their email reports look like: 2️⃣ LALIGA's reports have been accurate For every report we've received, we were able to verify that the URLs were hosting illegal streams of their copyrighted material. I have condemned LALIGA's unprecedented and indiscriminate blocks[2], and have warned of the potential for this power to be misused. So far, their reports have so far been valid. We expediently acted on them, in order to minimize the collateral damage. 3️⃣ Blocking hostnames vs blocking networks If you look at their email report above, you'll notice they single out an IP address. The crux of the issue is that in modern CDN networks, that IP address can represent hundreds or thousands of legitimate customers. The appropriate response would be to block *only the infringing hostname* by using the SNI fragment of the TLS handshake (e.g.: imagine blocking "𝚏𝚛𝚎𝚎𝚕𝚊𝚕𝚒𝚐𝚊𝚜𝚝𝚛𝚎𝚊𝚖.𝚝𝚟"). Since some CDNs don't offer this "granular blocking" possibility (given they encrypt SNI via a TLS protocol extension called "Encrypted Client Hello"), and ostensibly due to them not acting on the copyright reports, they're seeing significant collateral damage[3] With over 150,000 paying teams and thousands of Enterprise accounts hosting critical services in areas like health care, emergency response, banking, government, and more, we're always working to protect our uptime, security, and availability. [1] vercel.com/blog/update-on-sp… [2] xitter.cf/rauchg/status/19123164… [3] xitter.cf/dhh/status/19928429489…
Replying to @ajanerom
Spain blocks all of Cloudflare during those matches in an insane, medieval attempt to counter piracy. Totally batshit. (Omarchy is hosted with Cloudflare).
30
46
8
584
80,446
Oliver Montes retweeted
We're giving away an 11-page guide titled: “How to clone your voice and make an agent in <10 minutes” (Like we did for Elon and Karan). It can call support to complain on your behalf, make reservations, prank your friends, etc. You also get $100 in free credits so you can play with your voice AI. Retweet and comment "SONIC" below and we'll send you the step-by-step guide and $100 in credits.
536
307
39
468
89,355
Oliver Montes retweeted
🚨 There’s a large-scale supply chain attack in progress: the NPM account of a reputable developer has been compromised. The affected packages have already been downloaded over 1 billion times, meaning the entire JavaScript ecosystem may be at risk. The malicious payload works by silently swapping crypto addresses on the fly to steal funds. If you use a hardware wallet, pay attention to every transaction before signing and you're safe. If you don’t use a hardware wallet, refrain from making any on-chain transactions for now. It’s still unclear whether the attacker is also stealing seeds from software wallets directly at this stage. Excellent report here: jdstaerk.substack.com/p/we-j…
2,887
6,757
4,411
18,137
9,946,457
Oliver Montes retweeted
Now that the 2026 World Cup is closer, I've designed a more lifestyle shirt concept for El Salvador's Football National Team @LaSelecta_SLV
1
1
116
Oliver Montes retweeted
PHP server for Model Context Protocol, lets AI tools talk to your backend
1
2
13
3,259
Oliver Montes retweeted
Siento un poco de vértigo de poner una oferta laboral con un rango 50-60K. Hablamos que estará entre el 13,3% de los que más cobran en España. Para una micro empresa como la nuestra es tremendo. Al final nuestra apuesta es 1 en lugar de 2. Arriesgado pero veremos.
7
2
2
39
4,372
Oliver Montes retweeted
🚀Excited to announce F Lite: a new open-source text-to-image model by @freepik and @FAL! The first at this scale that’s both open-source and trained exclusively on licensed, high-quality data.🧵
8
48
15
258
161,101
Oliver Montes retweeted
GPT-4o está en otra liga. Los resultados parecen magia 🪄🤖
1
1
103