Permalink

In The Lord of the Rings, the palantír allows its user to see everything, but at a great cost: loss of control and manipulation. Palantir Technologies echoes this idea in the real world.

By aggregating massive amounts of data, Palantir provides governments and intelligence agencies with extraordinary surveillance and analytical power. While effective for security and intelligence, this concentration of data poses serious risks to privacy, transparency, and individual freedom.

As Tolkien warned, great vision without limits can be dangerous especially when few control what is seen.

palantir.com

Permalink

Sprites (fly.io)

Stateful sandbox environmentswith checkpoint & restore

A Sprite is a hardware-isolated execution environment for arbitrary code: a persistent Linux computer. Whether it's an AI agent like Claude Code or a binary your user just uploaded, Sprites are the simplest answer for "where should I run a blob of code".

The article technologically validates the choice of Incus/LXC. We are moving out of the era of the ‘everything-ephemeral’ mindset imposed by Docker and returning to stable working environments, which is essential for AI to be able to learn from its mistakes over time.

fly.io/blog/code-and-let-live

sprites.dev

sprites.dev/blog

demo

$ curl https://sprites.dev/install.sh | bash
$ sprite login
$ sprite create my-sprite
$ sprite console -s my-sprite
# claude --dangerously-skip-permissions (yolo)
 
Sprites execute code in Firecracker VMs. Even we have a hard time seeing what they're doing.
VMs run on isolated networks. Nothing can connect to your Sprite directly.
Post image
Post image
Post image
Post image
Post image
Post image
Permalink

Game interface for controlling agents

Post image
Permalink
Post image
Permalink
Post image
Post image
Post image
Permalink

I handle marketing for a 3D-printed Dr. Martens keychain
Photoroom + Affinity + CeWe print machine

Post image
Permalink

alien technology vera rubin

Post image
Permalink

LXC containers are quite interesting because they provide a real Linux OS with all UNIX tools available to agents like Opus 4.5, along with very fast machine (container) creation times. Kata also seems to be a good alternative for isolating agents.

Post image
Permalink

opus 4.5 x tmux = agentic farm

Post image
Permalink
Post image
Permalink

2026 will mark the moment when people truly begin using AI to create their own custom software.
Software is about to become radically personalized.

Post image
Permalink

Local AI

Ministral 3 
⌁ 3B, 8B, 14B
⌁ 256K tokens
⌁ Text + Vision
⌁ Apache 2.0

Versus

Beelink GTR9 Pro
128GB ram
2000 dollars

Post image
Permalink

Croc ! 🐊

I love this project written in Golang that makes it easy to transfer folders and files between two computers. I can send files from an LXC container behind CGNAT to Windows, macOS, or another Linux machine.

It uses a public relay that encrypts data in transit. By default, it uses the creator’s relay at croc.schollz.com.

You can also use your own relay; in fact, that could give the creator’s relay a break, which handles about 40 terabytes of bandwidth per month on Hetzner

https://github.com/schollz/croc

Post image
Permalink

~/.bashrc
alias cyolo='claude --dangerously-skip-permissions'
source ~/.bashrc
cyolo

Permalink

ia gen agentic coding

bash
grep
sed
awk
find
cat
head
tail
ls
cd
pwd

Permalink

Developers need computers that are always accessible online to run tasks or allow others to connect. These machines must be secure, with controlled SSH and web access. Setting up and maintaining complex authentication (passwords, OAuth, recovery flows, etc.) is a costly distraction from core work. 

Tape in your shell

⁂   ssh exe.dev
⁂   ssh exe.dev new --image=ghcr.io/marimo-team/marimo:latest-sql

⋌    https://nicolas.exe.xyz:8000

exe.dev is a great place to run a coding agent securely, with minimal supervision.
Kata Containers and Cloud Hypervisor used.

exe.dev

Post image
Post image
Post image
Permalink

Nico : AI systems are not deterministic like traditional software. Given the same input, they can produce different outputs. This probabilistic behavior can be unsettling for computer scientists and engineers who are used to strictly deterministic systems where the same code always produces the same result.

Post image
Close
Fullscreen image