Getting started
Sign in, make an API key, and launch, verify and delete your first confidential VM.
Confidential Cloud runs confidential virtual machines, called instances, on Intel TDX and AMD SEV-SNP hosts. The CPU encrypts each instance's memory with a key the host never sees, and every instance boots a measured image that you can check with a hardware quote before you trust it with anything.
You can drive it three ways, all on the same API and the same rules:
| Way in | Where | Best for |
|---|---|---|
| Console | console.confidential.ai | People: create and watch instances, manage your organization, API keys and billing |
| REST API | https://api.cloud.confidential.ai/v1 | Programs, scripts and CI |
| MCP server | https://api.cloud.confidential.ai/mcp | AI agents (see Connect an AI agent) |
This guide takes you from nothing to a running, verified instance in about five minutes.
1. Sign in to the console
Open console.confidential.ai and choose Sign in. Your organization's sign-in method appears on the next screen. If your account is not in an organization yet, the console asks you to name one, with Create organization, and makes you its owner. To join an existing organization instead, ask one of its owners to invite you from Settings > Organization.
Everything you create belongs to the organization you signed in to. Your role there decides what you can do:
| Role | Can |
|---|---|
viewer | See instances, reservations and metrics |
member | Everything a viewer can, and create, change and delete instances and request GPU reservations |
owner | Everything a member can, and invite people and approve GPU reservations that agents request |
2. Create an API key
Programs authenticate with an API key. In the console, open Settings > API keys and choose Create API key:
- Give the key a name, such as
laptoporci. - Pick its kind. Standard keys are for programs and CI. Agent keys are for AI agents that run without a person, and always expire (30 days by default, 90 at most).
- Grant only the permissions it needs. For this guide, tick View instances, Create and extend instances and Delete instances.
- Copy the key. It starts with
cc_and is shown only once.
export CC_API_KEY=cc_...Treat the key like a password. Anyone who holds it can act in your organization with the permissions you granted. Revoke it on the same page when you no longer need it.
3. Check what you can launch
Every request under /v1 sends the key and the API version:
curl https://api.cloud.confidential.ai/v1/images \
-H "Authorization: Bearer $CC_API_KEY" \
-H "Api-Version: 2026-10-15"GET /v1/images lists the images your organization can launch. Each one carries a launch contract: which disks to list, whether it takes SSH keys or a launch key, and which sizes it runs on under each TEE. This guide uses confos-cpu, the general-purpose image, which takes your SSH keys and needs one disk called user-data.
GET /v1/sizes and GET /v1/locations list the sizes (such as c1.2x: 2 vCPUs, 4 GiB) and the locations (such as us-west-1). See Instances for what each choice means.
4. Create an instance
curl https://api.cloud.confidential.ai/v1/instances \
-H "Authorization: Bearer $CC_API_KEY" \
-H "Api-Version: 2026-10-15" \
-H "Idempotency-Key: $(uuidgen)" \
-H "Content-Type: application/json" \
-d '{
"name": "hello",
"location": "us-west-1",
"size": "c1.2x",
"image": "confos-cpu",
"tee": "tdx",
"disks": [ { "name": "user-data", "type": "scratch", "sizeGiB": 10 } ],
"network": { "ingress": [ { "name": "ssh", "port": 22 } ] },
"access": { "sshKeys": [ "'"$(cat ~/.ssh/id_ed25519.pub)"'" ] }
}'The API answers 202 Accepted at once. The instance is provisioning, and its endpoints are already assigned:
{
"items": [
{
"id": "i-9pj5dy9i6cbu",
"name": "hello",
"status": "provisioning",
"statusReason": { "code": "starting", "message": "The instance is placed and its host is starting it." },
"tee": "tdx",
"location": "us-west-1",
"size": "c1.2x",
"endpoints": [
{ "name": "ssh", "host": "51.81.170.114", "ipVersion": "ipv4", "port": 10443 },
{ "name": "attest", "host": "51.81.170.114", "ipVersion": "ipv4", "port": 5699 }
],
"attestation": {
"endpoint": "attest",
"referenceManifest": { "digest": "sha256:8cb2d4…", "uri": "https://ghcr.io/v2/confidential-dot-ai/confidential-os-builder/blobs/sha256:8cb2d4…" },
"launchBinding": { "file": "pubkey", "sha256": "7f97f3…", "sha384": "0d92dd…" }
},
"boot": { "count": 0 }
}
]
}On a shared IPv4 address the platform picks the public port, so SSH is on 51.81.170.114:10443 here, not on 22. The endpoint named attest serves fresh hardware quotes.
Send a new Idempotency-Key for every create, and the same key when you retry it. A retry never creates a second instance. If you lose a response, GET /v1/instances?idempotencyKey=<key> finds what the create made.
5. Wait until it is reachable
curl "https://api.cloud.confidential.ai/v1/instances/i-9pj5dy9i6cbu?waitFor=reachable&timeoutSeconds=60" \
-H "Authorization: Bearer $CC_API_KEY" \
-H "Api-Version: 2026-10-15"The call holds for up to 60 seconds and returns as soon as the instance is running with its Reachable condition True. Call it again until then, and stop if the status becomes failed, stopped or terminated. A small instance is usually reachable within a minute of the create.
6. Verify it, then connect
The API never tells you an instance is trustworthy: the hardware does. Before you send the instance a secret, ask its attest endpoint for a quote over a nonce of your own, check the quote against the image's signed reference manifest and your own SSH keys, and pin the SSH host keys the quote covers. Verify an instance walks through it with a complete script.
Then connect with the pinned keys only:
ssh -p 10443 \
-o UserKnownHostsFile=./i-9pj5dy9i6cbu.known_hosts \
-o HostKeyAlias=i-9pj5dy9i6cbu \
-o StrictHostKeyChecking=yes \
root@51.81.170.1147. Delete it
curl -X DELETE https://api.cloud.confidential.ai/v1/instances/i-9pj5dy9i6cbu \
-H "Authorization: Bearer $CC_API_KEY" \
-H "Api-Version: 2026-10-15"Deleting returns 202 with the instance terminating, then 200 with a terminated tombstone once its guest and disks are gone. Everything on its disks is destroyed, and it cannot be undone.
Next steps
- Instances: TEEs, sizes, images, disks, statuses and expiry.
- Networking: open and close ports, dedicated IPv4, custom domains.
- Verify an instance: the full attestation check.
- Metrics and Grafana: your instances' CPU, memory, disk and network.
- Connect an AI agent: the MCP server.
- Limits and errors: rate limits, retries and every error code.