Donato Capitella

I'm a Software Engineer and Ethical Hacker, but mostly a tech enthusiast who likes to discover how things work by breaking them apart.

[Donato Capitella @dcapitella]

Introducing spikee, an open-source toolkit for testing LLM applications against prompt injection attacks that can lead to exploitation—like data exfiltration, XSS, and resource exhaustion. Easily create custom datasets to match your specific use cases.

https://t.co/It21KKWmJD

[Donato Capitella @dcapitella]

@antigravity Gemini 3.5 Flash is one of the biggest quality regressions I have ever experienced, easily gets into loops, hallucinates 8/10 times and breaks stuff.

[Framework @FrameworkPuter]

We spent the last few months overhauling our logistics infrastructure around refurbishment, and today we launched a broad set of refurbished products into the Framework Outlet. We'll be able to turn around customer returns into refurbs faster now too!

[Donato Capitella @dcapitella]

@wunderwuzzi23 🤣

[Donato Capitella @dcapitella]

Intel Arc Pro B70 (32GB) for Local LLMs: llama.cpp (SYCL/Vulkan), vLLM (Intel LLM Scaler) Benchmarks

https://t.co/ISELXNekf1

[Donato Capitella @dcapitella]

@loganthorneloe True, but I have found it a bit treacherous, it comes up with quite destructive oversights and bugs, your level of attention needs to be higher to catch them

[Donato Capitella @dcapitella]

@Ysrthgrathe42 @TeksEdge @Kyuz0 @TeksEdge can you please update the tag to my handle on here, @dcapitella ?

[Donato Capitella @dcapitella]

@Ysrthgrathe42 @TeksEdge @Kyuz0 well, that's not my account... on here I am simply @dcapitella. You can find all the information on https://t.co/Vx7ypZVP0U.

[Donato Capitella @dcapitella]

@Flo408800717920 @TeksEdge @Kyuz0 Llama.cpp, deliberately with 1 concurrent request.

[Donato Capitella @dcapitella]

@know2whyi1ft @TeksEdge @Kyuz0 Not by a lot and not on all model architecture and quants. This is mostly a llama.cpp implementation difference, they don't have optimized kernels for ROCm for all architectures and quants.

[Donato Capitella @dcapitella]

@_mohansolo @antigravity Finally some communication!

[Donato Capitella @dcapitella]

@giovanniJurado You using my latest toolboxes? Have you got llama-cockpit?

[Donato Capitella @dcapitella]

@robzolkos I have just released a new video with pi performance on local ahrdware (AMD Strix Halo / R9700) with Qwen 3.6: https://t.co/2xEfEMMqnI, also including full SWE Verified Mini runs, for people to get an idea of speed of completion and accuracy of the quants we can run locally.

[Donato Capitella @dcapitella]

In this video, I look at what works when running coding agents locally on Strix Halo and Radeon R9700. I cover pi and opencode running Qwen 3.6, and SWE-bench mini benchmark results: https://t.co/KzG6REd6cP

[Donato Capitella @dcapitella]

@CACandChill Charge 10x what we do today to fix bugs :)

[Donato Capitella @dcapitella]

@MrOnyszko @antigravity It was cheap and good at the beginning, and I like the Windsurf feeling. Then the price went up and quotas became impossible to work with, plus they just don't have capacity even for paying users. I'll likely be moving away unless the new 3.5 flash model is good

[Donato Capitella @dcapitella]

@antigravity - already fully broken the second time I launch the new Antigravity 2.0. Login flow is totally broken.

[Donato Capitella @dcapitella]

@Bob_Welton I think so, but I'm not a Windows user, so I'd recommend searching for tutorials online.

[Donato Capitella @dcapitella]

https://t.co/KJkM76Yu5P

MTP can accelerate LLM inference 2x, especially for coding agents. This video covers what MTP is and the performance improvements you can expect for Qwen 3.6 on AMD Strix Halo & Dual Radeon 9700.

[Donato Capitella @dcapitella]

@baldwinblueprnt lol - nice joke :)

[Donato Capitella @dcapitella]

@antigravity How is Antigravity 2.0 missing a search feature for projects and conversation!?!?!?!?!

[Donato Capitella @dcapitella]

@menticapti @Hyperoptic I think they should rebrand

[Donato Capitella @dcapitella]

@Hyperoptic having another outage within 1 month ...