Outpost

Self-service provisioning for short-lived cloud environments. Request a dev or staging environment, get a running ECS service and RDS database a few minutes later, and let it tear itself down when the TTL runs out — no ticket, no waiting on ops.

Built as a full platform: a FastAPI backend, a Terraform + GitHub Actions provisioning pipeline, a React dashboard, and a CLI.

How it works

1. You ask for an environment (name, type, TTL) through the UI or CLI.
2. The API writes a PENDING row and dispatches a GitHub Actions workflow.
3. Terraform provisions a VPC subnet, security group, ECS service, and RDS instance, then calls back to the API with the outputs.
4. The environment moves to RUNNING. A runbook is generated automatically — connection info, log commands, TTL reminder.
5. When the TTL expires, it doesn't just vanish: there's a 24h grace period (EXPIRING) where nothing changes, then it's paused (PAUSED — ECS scaled to 0, RDS stopped, still reversible) before a final destroy a week later if nobody resumes it. Destroy manually at any point and it skips straight to teardown. Either way, Terraform tears everything down and the record is kept for the audit log.

Every AWS resource is tagged with env_id, team, and ttl, which is how cost tracking and targeted destroys work.

Stack

FastAPI · PostgreSQL · SQLAlchemy/Alembic · React + TypeScript + Tailwind · Terraform · GitHub Actions · AWS (ECS Fargate, RDS, Secrets Manager, CloudWatch)

Running it locally

cp .env.example .env   # fill in GitHub OAuth credentials
docker compose up

API comes up on :8000, UI on :3000. docker compose exec api alembic upgrade head to apply migrations.

The Terraform/AWS side has already been bootstrapped (S3 state bucket, DynamoDB lock table, shared ECS cluster, OIDC role) — see terraform/README.md if you're standing up a separate account.

CLI

pip install -e ./cli
outpost auth login
outpost teams list
outpost env create --name my-feature --team platform-team --type dev --ttl 24
outpost env list
outpost env pause <env-id>
outpost audit list

auth login opens your browser and finishes automatically once GitHub redirects back — no token to copy or paste. Running it over SSH? Use outpost auth login --manual instead. See cli/README.md for the full command reference.

Why it's built this way

A few decisions worth knowing before reading the code:

• Provisioning is async. POST /environments returns 202 immediately; Terraform runs in GitHub Actions and calls back when it's done. Terraform takes minutes, not milliseconds — blocking the request isn't an option.
• No static AWS credentials anywhere. GitHub Actions assumes an IAM role over OIDC. Nothing to rotate, nothing to leak.
• One Terraform workspace per environment, state keyed by env_id in S3. Destroying one environment can't touch another's resources.
• Environments are soft-deleted. The row, audit trail, and cost history survive destruction — you can still see what an environment cost and who killed it.

Status

Core platform (auth, RBAC, provisioning API, Terraform modules, GitHub Actions workflows, web UI) is built, tested, and deployed. The full lifecycle — provision, pause, resume, extend TTL, destroy — has been verified end-to-end against real AWS. The CLI is feature-complete (env, audit, teams, auth) and tested against a live Postgres-backed API instance; only PyPI/TestPyPI publishing is still outstanding.