Git for your test database. Snapshot it, break it, put it back in seconds.

Website: pt-perkasa-pilar-utama.github.io/testate

Testate runs beside the system under test, takes data-only snapshots of its databases, and puts any of them back on demand. One container, one volume, no account, no telemetry. Version 1.1.0.

The problem

Every QA team knows this chat:

"Could you wipe the transactions on UAT? I can't create another order." "Reset DEV to the seed again please, the master data is off."

So a developer stops, SSHes in, and resets a database by hand. Or writes a reset endpoint that must never ship, and sometimes ships. Meanwhile QA does not automate, because every automated run needs a clean database first. And nobody lets an AI agent near a shared database to debug anything, because one hallucinated DELETE is one too many.

What Testate does

flowchart LR
  take["Take a state"] --> break["Run the test, dirty the data"] --> back["Check the state out"] --> take

• Take a state. A data-only snapshot of every database in a project, at one moment, with a name and tags.
• Break it. Run the suite, click through the app, edit rows in the grid, load a fixture.
• Check it out. One click or one API call puts every database back and reports per database. A stash is taken first, so even that is reversible.
• See what moved. Diff two states, or a state against the live database, down to the changed cell.

Nothing gets added to your application. Testate is a separate service that talks to the databases directly.

Quick start

docker run -d --name testate -p 7378:7378 -v testate-data:/data \
  -e TESTATE_SECRETS_ACTIVE_KEY="$(openssl rand -base64 32)" \
  -e TESTATE_ADMIN_PASSWORD=change-me-now-1234 \
  ghcr.io/pt-perkasa-pilar-utama/testate:1.1.0

The same image is on Docker Hub as snowfluke/testate, under the same tags.

Testate dials the database from inside its container, so localhost there means Testate itself. A database in another container on the same machine is reached by its container name once both share a network: docker network create testate-net, then docker network connect testate-net testate and the same for the database container, and the host is that container's name with the engine's own port. A database running natively on the machine is host.docker.internal, which Docker on Linux defines only when the run command above carries --add-host=host.docker.internal:host-gateway.

Open http://localhost:7378, sign in as admin with that password, and change it when asked.

Then: add a database under Databases, take a state under States, break something, click Check out. Where to point the host is the one thing that catches people, so read Connecting a database before you add the first one.

For anything you rely on, use Compose and an .env file: Deployment plan. No Docker? There is one binary per platform.

A look around

States. A project opens here: every state, newest first, HEAD marked, Check out on each one.

The API, documented by the instance itself. Everything the dashboard does is a REST call, and the running instance serves its own reference at /api/v1/docs.

An agent, looking. Register the MCP endpoint with an agent token and Claude reads the guide, finds the database, and answers in rows. Masked columns stay masked.

Where it sits

flowchart LR
  subgraph server["Your test server: dev, SIT or UAT"]
    app["Your app"] -- "reads and writes" --> db[("Its databases")]
    t["Testate"] -- "snapshot, put it back" --> db
  end
  qa[Testers] --> t
  dev[Developers] --> t
  ci[CI/CD] --> t
  agent[AI agents] --> t

Testate is one container on the same server, or the same network, as the databases. Nothing installs inside your app. Remove the container and nothing remains but the volume.

Who it is for

• QA. The main user. Test freely, reset in seconds, ask nobody.
• Developers. Reproduce a bug from the exact state QA saw it in. Let an agent inspect the data without a VPN and without fear.
• Infra. One image, one volume, one reverse proxy line. No reset scripts to babysit.

What it works with

Every engine here is real, driven over its own protocol, and proven in a contract suite against a real server. S3-compatible means Amazon, Cloudflare R2, Google Cloud Storage, Backblaze B2, MinIO and anything else that speaks the protocol; the connecting guide has the endpoint for each.

Also in the box: a data grid with filters, keyset paging and write mode; a read-only SQL console with saved queries; CSV and XLSX import with a dry run; fixture extraction as SQL or JSON; three roles (Administrator, Tester, Guest); an audit log of every write.

Reset from a pipeline

Every screen sits on the REST API, so a pipeline resets a database with one call before a run:

curl -sf -X POST "$TESTATE_URL/api/v1/projects/shop/checkouts?wait=120" \
  -H "Authorization: Bearer $TESTATE_TOKEN" -H "Content-Type: application/json" \
  -d '{"state_name": "seeded-baseline"}' | jq -r '.data.job.status'

Gate the step on the job's status, not the HTTP code. The full pattern, tokens, idempotency and the reference are in CI/CD.

Let an agent look

Testate speaks MCP. An admin creates a token of kind agent, scoped to a project, with a role: Guest reads, Tester may also write to a sandbox, take a state and put one back. The token reaches /api/v1/mcp and nothing else.

claude mcp add --transport http testate https://testate.example.internal/api/v1/mcp \
  --header "Authorization: Bearer tst_YOUR_TOKEN"

The first tool the agent sees is help. Reads run in a read-only transaction, results are capped, column policies mask values before the agent sees them, and every call is audited. Everything else, including what happens when a token expires: Agent access.

Know the limits

Testate is honest about what it is. Read these before you install it.

One binary, no Docker

Every release carries a standalone executable for macOS on Apple silicon, Linux on x86-64 and Windows on x86-64, with the dashboard and the migrations inside it. Download from the releases page, then:

chmod +x testate-*-darwin-arm64          # macOS and Linux only
TESTATE_DATA_DIR=./testate-data \
TESTATE_SECRETS_ACTIVE_KEY="$(openssl rand -base64 32)" \
TESTATE_ADMIN_PASSWORD=change-me-now-1234 \
./testate-*-darwin-arm64

It listens on 7378 and keeps everything under TESTATE_DATA_DIR, which has to be set. Every other variable in deploy/.env.example applies the same way.

Verify a download

Every release is built by this repository's own workflow and leaves a trail you can check without trusting us: SLSA build provenance and a keyless Sigstore signature on the image and on every binary, plus a CycloneDX SBOM.

gh attestation verify oci://ghcr.io/pt-perkasa-pilar-utama/testate:1.1.0 \
  --repo PT-Perkasa-Pilar-Utama/testate
cosign verify ghcr.io/pt-perkasa-pilar-utama/testate:1.1.0 \
  --certificate-identity-regexp 'https://github.com/PT-Perkasa-Pilar-Utama/testate/' \
  --certificate-oidc-issuer https://token.actions.githubusercontent.com

Both commands hold for docker.io/snowfluke/testate:1.1.0 too: it is the same manifest, copied by digest and signed again where it lives.

For a binary, run the same two commands on the archive, with --bundle <archive>.sigstore.json for cosign; the release also carries testate-<version>.intoto.jsonl, the provenance statement for every archive, for a check without network access (gh attestation verify <archive> --bundle <that file>). A download that fails either check is not ours; SECURITY.md says how to report it.

Operate it

Upgrades, rolling back a failed migration, backups, and rotating the sealing key: Deployment plan and Key rotation. Health lives at /api/v1/health/live and /api/v1/health/ready, with no credential, for the probes.

Run it from source

bun install
cp apps/api/.env.example apps/api/.env      # set the key and the admin password
bun run dev                                 # API on :7378, dashboard on :7379

bun run reset:dev --yes --engines, then bun run seed:dev, then bun run dev: a clean dev instance holding the demo project the test suite uses, on the databases from deploy/compose.engines.yml, with the admin on the password from apps/api/.env. CLAUDE.md has the rest of the commands.

Docs

PRD · Security standards · Technical specs · API specs · Connecting a database · CI/CD · Agent access · Deployment plan · Key rotation · Security

License

MIT. Copyright PT. Perkasa Pilar Utama.