User Inference Keys
Let users bring their own Ando inference connection into your app.
User inference keys are how app builders support bring-your-own-inference with Ando. The user creates or chooses an Ando connection, reveals its Virtual Key, and adds that key to your app's Ando provider settings.
Your app should treat the key as the user's inference boundary. The key is not your platform key. It carries the user's Ando account, connection, model access, budget controls, and usage trail.
User setup flow
The user opens your provider settings and selects Ando as an OpenAI-compatible provider.
The user signs in to Ando, creates a connection for your app, and keeps inference optimization and spend controls in Ando.
The user reveals the key only when they are ready to connect your app.
Your backend calls GET /v1/models, stores the key securely, and marks Ando connected.
Header
Ando Virtual Keys are bearer credentials.
Authorization: Bearer sk-ando_...Use the key with:
GET https://inference.andoai.xyz/v1/modelsPOST https://inference.andoai.xyz/v1/chat/completions
Do not send a Virtual Key to the Tempo MPP endpoint. Tempo MPP uses
Authorization: Payment, not bearer auth.
Storage requirements
Encrypt the key or store it in a secret manager. Keep only a secret reference in normal app records.
Display a short redacted preview after save, never the full value.
Let users replace the key without deleting their workspace or provider settings.
Let users disconnect Ando immediately. Stop sending requests with that key after removal.
What to expect
- Missing or invalid keys return
401 Unauthorized. - Budget or spend-cap exhaustion can return
402 Payment Required. - Rate limits can return
429 Too Many Requests. - Ando usage remains visible in the user's Ando account and connection.
- Model availability should come from
GET /v1/models, not a hard-coded list hidden in your app.
Product guidance
When the user brings Ando, make the boundary visible.
- Use one Ando key per user, workspace, tenant, or project boundary.
- Name the provider connection in your app so the user can recognize it later.
- Include a
Test Ando connectionaction in settings. - Include a
Rotate keyaction next to the connected state. - Link budget and usage questions back to Ando instead of masking them as generic provider failures.
- Do not log Virtual Keys, prompt content, request bodies, response bodies, or model output in normal analytics.
Bring-your-own-inference works best when the user can see exactly where the model call is coming from, which account pays for it, and how to turn it off.