Quickstart

From request to first lease in five minutes.

Once your beta request is approved, the path from zero to a running workload is short. Here's the whole loop on one page; the deeper detail lives in the docs.

Hardware

What you need before you start.

Linux x86_64 or aarch64 on the host running the daemon.
An NVIDIA GPU (any modern data-center class) for GPU leases. CPU-only deploys also work.
Outbound HTTPS to the cell scheduler if you're using a managed cell.
For self-hosted, plan on at least one node for the daemon and one for the scheduler. They can be the same machine for a quick try.

Tenura also runs on Raspberry Pi 5 bare metal — useful for kicking the tires without a real GPU. See the Pi5 bringup guide.

Step 1

Request access.

If you haven't yet, submit the beta-request form. Approved teams receive a one-click signup link by email along with a beta token. We review weekly.

Request access

Step 2

Install the CLI.

The installer drops grafos on your PATH and registers your beta token.

curl -fsSL https://get.tenura.systems/install.sh | sh

Prefer to read the script first? It's published at the same URL — fetch and inspect before piping. Manual install steps are in the docs.

Step 3

Mint a tenant certificate.

mTLS is the floor for the control plane, so every tenant needs a certificate signed by the fabric CA. For managed cells, we issue this for you and the installer pulls it down. For self-hosted, you bootstrap your own CA and mint a tenant cert with one command:

grafos admin init-ca --out ./ca
grafos admin issue-cert --ca-cert ca/ca.pem --ca-key ca/ca-key.pem \
  --tenant acme --out ./acme

Keep the CA key offline. Distribute the tenant cert + key to the people authorized to deploy under that tenant.

Step 4

Deploy your first workload.

Build a WASM program (the docs have a hello-world template) and deploy it. The scheduler admits the request, mints a capability token, and returns a lease ID:

grafos deploy run inference.wasm \
  --scheduler https://cell.us-west.tenura.systems \
  --tenant acme --gpu 1 --mem 80G --ttl 5m \
  --cert acme/client.pem --key acme/client-key.pem --ca ca/ca.pem

The CLI prints the lease ID, the bound resources, and the time-to-expiry.

Step 5

Verify the lease.

List active leases and watch lifecycle events stream:

grafos lease list --tenant acme
grafos lease watch --tenant acme

When the TTL elapses (or you call grafos lease release), teardown runs and the resource returns to the pool. Confirm with grafos lease list.

What's next

Where to go from here.

The cookbook has end-to-end recipes for inference, training, pipeline jobs, and multi-cell deploys. The architecture page covers the programming model: leases, capabilities, collections, placement, and recovery. The economics page covers how leases change capacity planning and billing.

Stuck? Email beta@tenura.systems with your tenant ID and we'll help.