STEP 01
One command
You or your agent type the same command. --json returns machine-readable output, --wait waits until the deploy finishes, --dry-run checks before anything runs.
- The API key is read from
TOSE_API_KEY (starts with tose_sk_) or --api-key-stdin. tose doctor and tose capabilities let an agent check its environment before it starts.
STEP 02
Build
Connect a repository. With a Dockerfile, TOSE builds from it; without one, TOSE generates it for common runtimes: Node, Bun, Deno, Python, Go, Rust, Java, PHP, .NET or a static site.
- linux/amd64 builds, 2 GB memory, 45-minute limit. A new deploy is refused (HTTP 402) when the wallet is empty.
STEP 03
Run
The app runs on Kubernetes with a domain and SSL set up for it. Add a custom domain and TOSE issues and renews the certificate. Change vCPU, RAM and replicas when you need to.
- A new project starts at 1 vCPU, 2 GB RAM and 1 replica.
- Managed PostgreSQL, MySQL, Redis and MongoDB; persistent disks keep data while a workload is stopped.
STEP 04
Humans stay in control
Agents work within the scope you grant. Each workspace has one prepaid wallet, and it is a hard spending limit: an agent cannot spend more than the balance. If something goes wrong, roll back in one click.
- Scoped API keys (Beta): The dashboard form and the CLI cannot set scopes yet; a key without scopes keeps full workspace access.
- Out of balance: compute pauses and disks are kept; top up and the project starts again.