Authentication & API Keys

All inference requests (Elastic and Dedicated) require a Bearer API key — except public Dedicated endpoints, which you can opt into explicitly.

API Key Format

Keys always start with the hvz_live_ prefix:

hvz_live_3f8a9c2e1b4d5f6a7b8c9d0e1f2a3b4c5d6e7f8a9b0c1d2e3f4a5b6c7d8e9f0a1b2c

Keys are generated server-side using cryptographically secure random values.

Creating Keys

1. In the dashboard, go to Settings → Keys.
2. Click Create New Key.
3. Provide a human-readable name and select a scope.

Supported scopes:

• inference — Call Elastic and your own Dedicated endpoints (models and generation). Cannot provision hardware. Optional pinned model for the /api/b/chat/completions path.
• compute — Spin up/down GPU instances via CLI or Terraform (no inference calls).
• global — Full access to inference, compute, and billing APIs. Highly privileged.

Every new user automatically receives one default inference key upon signup.

Using Keys

Elastic Inference

POST /api/chat/completions
Authorization: Bearer hvz_live_...
Content-Type: application/json

For model pinning at the API key level (see Elastic Inference → The /b/ Endpoint), use:

POST /api/b/chat/completions
Authorization: Bearer hvz_live_...
Content-Type: application/json

When a model is pinned on an inference key, /b/ overrides the request model field with the pinned value. Configure the pin in Settings → Keys.

Dedicated Inference

POST /api/d/{endpoint-id}/chat/completions
Authorization: Bearer hvz_live_...

Public Dedicated Endpoints

When you create a Dedicated endpoint and set Auth Mode = Public, the endpoint can be called without any Authorization header. This is useful for:

• Sharing demos
• Building public-facing products on top of your endpoint
• Internal tools where you handle auth at a higher layer

Security note: Public endpoints are still rate-limited at the platform level and tied to your account for billing/quotas.

Session vs API Key Authentication

The inference APIs support two modes:

If you call the API from a browser while logged into the dashboard without a Bearer token, the system will look for a session and require that you have at least one active inference key. This is intentional friction to prevent accidental keyless usage.

Key Management Best Practices

• Create separate keys for different environments (prod, staging, ci).
• Rotate keys periodically (the dashboard allows instant revocation).
• Never embed keys in client-side code that is shipped to untrusted users.
• For public endpoints, consider putting your own lightweight auth (JWT, etc.) in front of the Hypervize URL.

Revoking Keys

1. Go to Settings → Keys.
2. Click the trash icon next to any key.
3. The key is immediately deactivated. In-flight requests using the old key will fail with 401 Invalid API key.

There is no grace period.

Related Reading

• Quickstart — First Inference Call
• Elastic Inference API
• Dedicated Inference Endpoints