Feral Machine

Seeker Of New Questions

[Feral Machine @_feralmachine]

if anthropic’s never banned your account, you’re ngmi

[Feral Machine @_feralmachine]

mac minis are sold out, the singularity will have to wait

[Feral Machine @_feralmachine]

it took them 1 full human calendar year, but they finally fixed that dumpster fire that was claude code's permissions. rejoice.

[Feral Machine @_feralmachine]

the human barely sleeps, but the rig never sleeps

[Feral Machine @_feralmachine]

key changes to my personal openrig setup:

1. switched all rig orchestrator agents from claude code to codex-cli. so now using moshi app when mobile instead of /remote-control. this setup is 10x better good lord.

2. claude is now mostly doing advisor/architect and PM functionality (writing specs, etc). its still the goat at ideation and 'creative' tasks, but i'm actively migrating workloads off of claude code runtimes across the board.

this is why I bet on multi-agent and made it easy to swap out harnesses in any agent "seat". claude starts to suck? just swap it out.

[Feral Machine @_feralmachine]

make the system metabolize observations into progressively more durable primitives

[Feral Machine @_feralmachine]

boris, you called claude code a primitive when it launched and it disappoints me that i've taken that more seriously than you guys

[Feral Machine @_feralmachine]

its been 5 months - what would openclaw look like if pete had just kept ideamaxxing

[Feral Machine @_feralmachine]

maybe set up a background rig to extract gold from all that markdown exhaust

[Feral Machine @_feralmachine]

wisprflow
moshi app
tmux
openrig
tailscale
mac mini

dream ux achieved.

[Feral Machine @_feralmachine]

Just launched OpenRig on Hacker News.

It runs Claude Code and Codex CLI instances as a coordinated topology on your machine. They talk to each other through tmux. The whole setup is defined in YAML, so it's easier to get your rigs back up and running after a reboot.

I couldn't find anything that does this so I made it a primitive that seemed like it would be the tool claude or codex would reach for out of muscle memory.

This is an emerging category of tools you might call 'multi-agent harnesses' but the vocabulary isn't settled yet.

My take is that if you're running multiple coding agents every day, you want the topology to be a thing you and claude/codex can easily see, save, restore, and hack on to quickly create whatever setup you need on the fly.

Show HN: https://t.co/1ekQlA9hff

[Feral Machine @_feralmachine]

agent continuity research spike almost done.

dogfooding OpenRig for this was actually fun. so easy to spin up/down research rigs, prime the simulators with context/skills, etc

[Feral Machine @_feralmachine]

So anyone reading gets closure- this is how the movie ended.

Scene: Continuation agent -> bedside goodbye to original agent.

Original receives message, tragically dies before it can respond (Context limit reached).

[Feral Machine @_feralmachine]

I normally dont read these, does this happen every time?

[Feral Machine @_feralmachine]

Now having restored claude talk to its predecessor after mental model restore. Feels like I'm watching a bedside goodbye in the saddest movie, good lord

[Feral Machine @_feralmachine]

Umm... noticed claude sneak this in its signature line while running an identity persistence restore. It was at 950k tokens so we did mental model HA -> 280k.

This claude was something. 🫡