everyone’s running 5 coding agents now. somehow configuring localhost is still my job.
separate branches, but they’re fighting over the same ports and database.
so i built isola. each git worktree gets its own ports, localhost urls and postgres database.
https://t.co/pqHNJDPIE4
@Armandotrue makes sense. the part that broke for me was the step after building, every worktree wanting the same port and the same db. ended up building isola for it
@levelsio worth knowing if you use ufw on the box instead: docker writes its own iptables rules, so published container ports skip ufw completely. the hetzner cloud firewall sits outside the vm so it doesnt have that problem
@ea_vlad_knysh cookies ignoring ports is such a sneaky one. isola does the same per worktree, branch.project.localhost. the per account part is clever though, two logged in users without an incognito window
@ctatedev nice. can you run one instance per worktree with its own state and port? parallel agents each getting their own fake cognito and slack would be great
@IntCyberDigest also check the clis that ship their own runtime. claude code bundles bun, so updating node on your machine doesnt patch it. claude update does
@BHolmesDev how do you handle secrets in them? agents cat .env files and tokens land in tool output, so a shared store is also a shared secrets dump. scrub on upload?