A separate local project per coding agent is a quiet but big deal for agent testing: reproducible state on every run. Does the per-agent reset also cover auth and storage state, or just the database?

does each agent project get its own postgres port

Finally. Docker was the heaviest part of local dev for me. A separate local project for each agent is a smart touch too, since parallel test runs stomping on one DB was a real pain.

Per-agent local projects are the part I want. Most flaky agent runs I debug trace back to two agents sharing one database. One resets state mid-task, the other fails, and it looks like a model problem. Isolated DBs with a seed snapshot per agent fix that.

Life changing, it would be perfect if it had a bit more support for WSL.

isolation per agent is the actual win here

Does the per-agent isolation cover local Storage files as well as the database? Upload tests are where I'd be most worried about two agents getting in each other's way.

separate git branches don't help much if both agents are resetting the same database lol. the per-agent project is the part i'd want to try.

Pglite but better?

supabase start used to need docker to run the whole stack in containers, postgres, auth, storage and the rest, so dropping that makes local setup a lot lighter

That's awesome! It's great that you're listening to user feedback and making Supabase even more accessible.

Massive! We can now test our migrations isolated when working with multi agents

Nice to have a separate database per worktree. How do you catch migration conflicts when two agents change the schema independently and their branches get merged?

every agent getting its own local project is the real headline. no more agents stepping on each other's db

"Decoupling dev environments from Docker simplifies onboarding. The real win is enabling isolated database states for parallel testing, which is critical for complex integration suites

separate local projects per agent is the bit i care about most. one agent resetting the db under another one's test run is such a sneaky way to get flaky failures.

Per-agent local projects is the part I like. One agent resetting its database while another is mid test run is exactly the collision that makes running parallel agents messy.

A project per agent is the unglamorous fix parallel coding agents actually needed. Two agents sharing one dev database is how you get flaky tests nobody can reproduce.

per-agent projects need deterministic port allocation or parallel runs still collide

Project per worktree is the bit I'd actually use. Parallel agents sharing one local db is exactly where they trip over each other, one runs db reset while the other is mid migration. Has anyone tried the native runtime inside Codex cloud envs yet?

without docker is the headline. the real unlock is each agent owning its own local project so one reset doesn't kill the other run

Isolating coding agent execution tracks by deploying per-agent local Postgres projects without container overhead is a massive paradigm shift for AI infrastructure, but many platform teams misconfigure the lifecycle boundaries. Chaining parallel database resets mid-task without a dedicated transactional seed snapshot layer guarantees subtle dynamic state contamination when separate agents step on shared connection configurations under heavy load testing. Look at the isolated database pool profiler below. How does your backend handle session state virtualization to prevent agent test failures? Let's discuss! 👇

Giving each agent its own project is the part I care about. Two agents sharing one test DB is how you get flaky CI nobody can reproduce.

Ditching the Docker daemon overhead is such a massive quality of life upgrade for quick local test loops.

no docker is the part I like. per-agent project so one resets while another is mid-run is the real unlock though — no shared state fights. running this local tonight.

fast enough to spin up per test run?

finally no docker for local supabase

Love the per-agent local projects. One trap when those agents ship: the local service_role habit ends up in the client bundle. Keep RLS on every user table before deploy, even if local felt open.

One agent resetting the db while another is mid test run is exactly the kind of flaky mess this saves. Nice

no Docker for local Supabase is the real win here. the per-agent local project bit is what I actually needed - two watchers resetting the same DB mid-run was my whole week

per-agent local project is the detail that matters. one agent resetting db while another is mid test run was always the pain. docker requirement gone helps too

The per-agent local project is the part that matters. Shared local DBs are how one agent's cleanup wipes another agent's test state mid-run. Do you wipe the project when the agent session ends, or keep it around for the next run on the same thread?

parallel agents sharing one local db always ended with one agent debugging its own code because another had just wiped the tables

Too late brother, Ive moved ro planetscale. Your query is slow as fuck