Ghost for developers — Unlimited Postgres for your agents

Source: original

Your agent needs more than 2 projects.

Most Postgres providers cap you at a handful of projects. Agents blow through that in a single task. Ghost gives your agent unlimited databases. Fork, test, throw away.

curl -fsSL https://install.ghost.build | sh

Weekly digest

curl -fsSL https://install.ghost.build | sh

Subscribe to the weekly digest →

Send

Unlimited databases·Unlimited forks·100 hrs/mo free·1TB storage free

six ways in

Ghost is unlimited, but here are six ways to use it.

When setting up a database is a two-minute side quest, you only do it when you have to. When it's free and instant, you start using forks like scratch paper.

01

Let Claude touch your database

Fork main before you let your agent run. If it does something weird, drop the fork. If the migration actually worked, keep it. Your real data stays untouched.

→safe vibe-coding with a real db

02

Try two schemas, keep one

Not sure if you need a separate users table or if you can just add columns? Fork it twice. Build both. Ship the one that feels right. Delete the other.

→decisions without commitment

03

A scratch db for every prompt

Prototyping with an LLM? Spin up a throwaway db for each idea. Load sample data. Try stuff. Close the tab when you're done. No cleanup.

→no commitment

04

Undo a bad migration

Ran a migration that broke prod at 11pm? Fork from an hour ago. Point your app at the fork. Debug in the morning. Your users never notice.

→lower-stakes mistakes

05

A real db for every PR

Your collaborator opens a PR with a schema change. Fork main, apply the migration, check that everything still works. No "can you push to staging" slack messages.

→preview envs that actually work

06

Test locally against real data

Your tests need realistic data but you're tired of seeding fixtures. Fork from prod (or a sanitized copy). Run your tests. Drop the fork. Repeat.

→no more fake data

a real experiment

Fork, run, decide.

Four phases. Each completes in milliseconds. The loop closes faster than most people's postgres clients connect.

01 · setup

Fork three ways

Start from a source database. Create three independent forks, one per hypothesis.

+184ms total

02 · run

Experiment in parallel

Each fork runs in its own isolated Postgres. Nothing leaks between them.

minutes, not hours

03 · compare

Measure against the same baseline

All three forks share a common ancestor, so results are comparable out of the box.

on your schedule

04 · decide

Keep one. Drop the rest.

Promote the winner. Discard the rest. No cleanup, no manual teardown, no ghosted resources.

39ms per drop

see it in code

An embedding bakeoff, start to finish.

Fork the source three ways, run the evals in parallel, keep the winner. Three commands, real CLI grammar, no cleanup step.

EMBEDDING BAKEOFF · three models, one source

done

fork the source database three ways

ghost fork docs-main --name eval-small

58ms

ghost fork docs-main --name eval-medium

61ms

ghost fork docs-main --name eval-large

64ms

run the eval on each fork in parallel

eval-small: all-minilm · recall 0.71

eval-medium: bge-base · recall 0.84

eval-large: voyage-3 · recall 0.92 ✓

keep the winner, delete the rest

ghost delete eval-small --confirm

39ms

ghost delete eval-medium --confirm

41ms

three forks · two deleted, one kept

built with ghost

What people are making.

A few projects built on Ghost by developers and vibe-coders. More on the gallery.

pacmandb @scott Pac-Man, but every pellet eaten is a row in Postgres. Ghosts are forks. Ghost City @jacky A multiplayer city-builder where every player gets their own database. Fork to clone a city. March Madness @ssilv Bracket predictor that forks the tournament state on every upset. Try every timeline.

Stop rationing your experiments.

If you've been running one careful test at a time because setup is expensive, that constraint is gone. Fork freely. Try everything. Keep what works.

curl -fsSL https://install.ghost.build | sh

read the forking docs →