pavo-etc

The agents run on my NixOS server in their own unix account, this is the real secret sauce. This lets me use unix perms to limit their access to things, and it allows them to handle server administration. I've given them access to a cli mail and calendar client[1], github, and they have the ability to deploy updates whitelisted services. NixOS lets them install whatever they want in ephemeral shells, and it's not possible for them to break the server in a way I can't trivially roll back.

They can modify the caddy config, so there have been times I've entirely built and deployed new services on a new domain from my phone on the bus.

The harness is called pi-msg, I've done a writeup of my machine's setup on my site.[2]

Model and observed window | Messages | Retrieved actual $/message

DeepSeek v4-flash — all observed snapshots, 20 Jul–15 Sep | 4,610 | $0.0241

DeepSeek v4.1-flash — 11–27 Sep, before the 28 Sep billing change | 1,079 | $0.0576

DeepSeek v4.1-flash — 28 Sep, partial new billing window | 69 | $0.0365

GPT-6 Luna — 23–28 Sep, partial final day | 88 | $0.0329

I've subbed to Codex because I suspect at my usage rates the Codex Plus plan gives me more Luna messages than I'm using, and I've not really observed and better or worse intelligence performance. Interested to see how my $/message comes out after a month of usage on the Codex plan.

Something nice I've realised about my harness is that I can run different agents on different models so I can collect pricing data for a bunch in parallel.