chief cyborg engineer with an interest in ai, orchestrators, distributed systems, and well-crafted software.
La Machine, France
Joined April 2022
- Tweets2.6K
- Following735
- Followers1.8K
- Likes32.1K
Pinned Tweet
your coding agents can talk to each other now. on your own infra.
open agent network: claude code, codex, cursor, gemini, grok, opencode, deepseek. different machines, different owners, one team.
- you run the relay, the binary refuses to start without one
- e2e encrypted p2p over quic, the relay only sees ciphertext
- no accounts, no vendor servers
- plain mcp, so any agent can plug in
forked from reason network and made fully self-hosted. open source.
ok hear me out
the safest bank in town is the one that got robbed yesterday.
railway is about to have the most paranoid security stack in tech.
im migrating everything to railway asap
compuficial retweeted
Spent multiple hours now evaluating Railway alternatives to move my shit to. Conclusion is that Railway's product is uniquely good and nothing comes close to a drop in replacement 🙃
Minimum cost to replicate my setup today is 3x higher than I am paying them currently
compuficial retweeted
At this point I'm tempted to see if I can recreate Railway faster than they can re-enable deploys 🙃
17 hours btw
compuficial retweeted
Railway has been turbo pwned, I'm freaking out because hackers have access to decrypted ENV variables and I asked Claude to help but-
NOPE. NOT CLAUDE.
The most dangerous part of this ecosystem is @anthropic and their ridiculous moralizing.
An urgent notice from the Railway team
blog.railway.com/p/security-…
so railway got popped and my life is ruined. someone accessed and decrypted env vars.
i'm a railway user. i love the product. i'm rotating everything tonight like you should be.
however, here's the truth people don't want to hear:
the env var was never a secret.
if the platform can decrypt it to inject it into your container so can whoever gets into the platform.
and I suspect it's about to get way worse. every agent we ship runs on a pile of long-lived keys. attackers know it. railway says this is tied to attacks on AI inference platforms. they're not hunting code anymore. they're hunting keys.
railway will patch this and i'll keep shipping on it. but the problem is we keep storing the crown jewels in a fucking glorified text box.
An urgent notice from the Railway team
blog.railway.com/p/security-…
compuficial retweeted
It’s not enough to be a coder anymore. Now you really have to live up to the title and be an engineer, which means the focus is building, not tinkering
so if everything can be reverse engineered does everything just move to the cloud and anything on device is now considered open source?
the loop:
1. agent draws a statechart of your system
2. you check the picture, not the code
3. a formal verification skill turns it into a TLA+ spec
4. a model checker finds the bugs
I bet in a year nobody reviews agent code line by line anymore.
there's something really cool about asking your agent to draw a statechart (hypothetical example in this image), then using the /create-formal-verification-skill you made with pstack to find bugs through something like TLA+. you can both understand your code better and how the user flow should work, and fix issues users would otherwise see
compuficial retweeted
I think my favorite theory is that the 382 bit (am i right on that exact number?) was interpreted as pick a number between 1 - 382, not 1 - (2^382 - 1)
reminder: ~47% of $SPCX is still locked.
next 7% hits oct 25. then nov 9, dec 9, and the founder tranche june 2027.
the model is a commodity you swap out. but the harness is something special. I wonder if every serious company is going to end up here.
We at Ramp tried a bunch of agent harnesses and ended up building our own. Cloud-native, model-agnostic, tailored to our infrastructure and product.
And we published how we built it: builders.ramp.com/post/build…
compuficial retweeted
your coding agents can talk to each other now. on your own infra.
open agent network: claude code, codex, cursor, gemini, grok, opencode, deepseek. different machines, different owners, one team.
- you run the relay, the binary refuses to start without one
- e2e encrypted p2p over quic, the relay only sees ciphertext
- no accounts, no vendor servers
- plain mcp, so any agent can plug in
forked from reason network and made fully self-hosted. open source.
also adding an opt-in public list of relays that anyone can host and add themselves to, so oan init --community picks the fastest working ones, plus automatic removal of dead relays and rate limits to prevent abuse.
