Gabriel Fontes
d5c0787ba0
The Cortex-A72 in the BCM2711 implements no ARMv8 crypto extensions, so TLS runs entirely in software. Measured on merope: AES-256-GCM manages 53 MB/s per core against 208 MB/s for ChaCha20-Poly1305, roughly 4x. All six providers default to TLS_AES_256_GCM_SHA384 and all six accept ChaCha20, so at the 43 MB/s this box was pulling, about 0.8 of a core was going to decryption alone -- on a machine already sitting at 93% CPU with the network only a third utilised and the disks coasting. Usenet throughput here is bounded by the cipher, not by bandwidth or IO. Setting ssl_ciphers makes SABnzbd cap the connection at TLS 1.2, since Python never exposed SSL_CTX_set_ciphersuites() for the 1.3 suites. That costs an extra handshake round trip, amortised across long-lived connections, and an unencrypted certificate that SNI already gives away. ECDHE preserves forward secrecy and ChaCha20-Poly1305 is the same AEAD 1.3 would have picked. Preferred over exporting OPENSSL_CONF, which reaches the same cipher while keeping TLS 1.3: this is the supported knob, so when upstream can select 1.3 ciphersuites the cap lifts on its own, whereas overriding a library's config file to reorder one list is the kind of thing that breaks silently. Assisted-by: pi (claude-opus-5)
Foundry
My public infrastructure monorepo: the NixOS/home-manager configuration that runs all my machines, plus the source of the personal projects it deploys. Requires Nix flakes.
Looking for something simpler to start out with flakes? Try my starter config repo.
Repository layout
hosts/nixos/ per-machine NixOS configurations (atlas, alcyone, ...)
hosts/system-manager/ configurations for non-NixOS Linux systems
home/ home-manager configuration (feature-flagged)
modules/ reusable nixos/ and home-manager/ modules
overlays/, pkgs/ package overlays and custom packages (including wallpapers)
lib/ pure-Nix Material You color engine
projects/ my public projects deployed from here (e.g. the m7.rs website)
Highlights:
- NixOS configurations: desktop, laptop, servers
- Opt-in persistence through impermanence + blank snapshotting
- Encrypted single BTRFS partition (with disko for declarative partitioning)
- Secure Boot via lanzaboote
- Fully declarative self-hosted stuff
- Deployment secrets using sops-nix
- Mesh networked hosts with tailscale and headscale
- Flexible Home Manager configs through feature flags
- Extensively configured hyprland environment
- Declarative theming: wallpapers and a pure-Nix Material You color engine
- Hydra CI/CD builds every host, serves a binary cache, and hosts auto-upgrade by pull deployment
About the installation
All my computers use a single btrfs (encrypted on all except headless systems)
partition, with subvolumes for /nix, a /persist directory (which I opt in
using impermanence), swap file, and a root subvolume (cleared on every boot).
Home-manager is used as a NixOS module, integrated via home-manager.users.
Secrets
For deployment secrets (such as user passwords and server service secrets), I'm
using the awesome sops-nix. All secrets
are encrypted with my personal PGP key (stored on a YubiKey), as well as the
relevant systems' SSH host keys.
On my desktop and laptop, I use pass for managing passwords, which are
encrypted using (you bet) my PGP key. This same key is also used for mail
signing, as well as for SSH'ing around.
Tooling and applications I use
Most relevant user apps daily drivers:
- hyprland + hypridle + hyprlock
- waybar
- helix
- fish
- alacritty
- qutebrowser
- neomutt + mbsync
- khal + khard + todoman + vdirsyncer
- gpg + pass
- tailscale
- podman
- zathura
- wofi
- bat + fd + rg
- kdeconnect
Some of the services I host:
- hydra
- jellyfin
- *arrs (including torrent and usenet)
- prometheus
- websites (such as https://m7.rs)
- minecraft
- headscale
Nixy stuff:
- sops-nix
- impermanence
- disko
- lanzaboote
- home-manager
- and NixOS and nix itself, of course :)
Let me know if you have any questions about them :)
Unixpornish stuff
AI usage note
Since June 2026, I've been trying out LLM assistance in my workflows. So far it feels pretty good; brainstorming helps me a lot with decision paralysis. I'm trying to keep my use bounded and disclosed. I think it's a useful tool, but it should be adopted with care.
I will write a decent blog post about my opinions on AI at some point. The gist is:
- These things are useful for bounded tasks with good docs and reviews, but suck at owning architecture or accountability
- There's no going back now that open-weight models run on consumer hardware, we can't uninvent them
- Boycotting AI does not help by itself; non-use is not a political strategy
- We need clear LLM policy on projects rather than trying (and failing) to forbid it
- Disclosure is very important, and trying to pass LLM output off as human-written is disrespectful
- LLMs should never have been built by scraping and exploiting our art and work
- If copyright doesn't protect our creations, it shouldn't protect their models; support open-weight models and distillation as harm reduction
- We need regulation, decentralization, and redistribution of the value LLMs generate
- Machines can't be horny, thus can't create art; LLMs are for utility, not art or craft. Pay an artist instead
Some bibliography I'd recommend:
- Alice's "AI Sucks. Hating it is not enough." (please read it, it's an amazing piece)
- Drew's "The cults of TDD and GenAI" (I don't really agree with the non-use perspective, but otherwise I think it's solid)
- Armin's "Communities of Not" and "The Center Has a Bias" (I'm not a centrist politically, though)