Skip to content
Documentation menu

Help

One request, one route.

Your app calls the Ompera API. Ompera checks the key, finds the model you named and passes the request to the device that runs it. The answer comes back the same way.

The route

  1. Your app

    Sends an OpenAI-style request to https://api.ompera.net/v1 with your API key and a model address.

  2. Ompera API

    Checks the key, resolves the address to one connected model and sends the request over an encrypted connection. It never stores the content.

  3. Your device

    For a computer: the Ompera tab in Chrome passes the request to your model server on localhost through the extension. For a phone: the Ompera app runs the model itself.

Your computer connects out to Ompera, so there are no ports to open and no public IP to set up. If the device goes offline, requests fail with a clear error. Ompera sends them to a different model only if your key is set to use a pool model when yours is offline, and the response names the model that answered.

Model addresses

AddressWhat it isWho can use it
my/localA model connected from your computer.You, with a key that allows Unshielded models.
HomeLab/qwen3-4bA model a member of your pool runs. If several members run it, the free copy answers.Everyone in the pool.
android-qwen3-…A network model running in the Ompera Android app.Everyone on Ompera.
my/qwen3_4b_instructA model the Ompera app runs itself and keeps to its owner (Me and my pools).You, and your pool members at <pool>/<model>.

Ask GET /v1/models for the exact addresses your key can use.

Services

Translation, text from images, speech to text and embeddings are services, not models, so your app calls the service's endpoint. Embeddings also name a fixed engine for a compatible search index, and Ompera sends the request to a free provider that turned the service on and passed a test request for it. Each service runs on an engine built for that task, never on a chat model. Services API →

Which sign-in or key goes where

CredentialUsed forWhere it lives
Passkey or walletSigning in to your workspace and the connector page.Your device or password manager.
Ompera API keyYour apps' AI requests. Cannot open your workspace.Your app's settings or environment.
Server API tokenOnly if your local model server requires one.The connector's Advanced settings.
Phone pairing codeLinking the Android app to your account, once.Shown in Workspace → Devices for 10 minutes.

Pools

In a pool, another member's request takes the same route to your computer. Ompera checks that they are in the pool; they can use the models you connect, but never your keys, your workspace or anything else on your computer. How pools work →

When the model is busy

A computer-connected model answers one request at a time. Extra requests wait in line for up to 30 seconds, then receive model_busy so your app can retry.

The rating of a shared model

If you share a model, Your sharing record in Workspace → Devices shows a Rating from 0 to 100. It says how dependably your devices answered in the last 30 days. During the beta it is information for you: it does not change who can use your model.

  • It starts from the share of requests your devices answered, your own requests included. A request the caller cancelled, or one the model refused as it was written, counts neither way.
  • A request counts against a device when it is cut off while being answered, when the model server returns an error, or when the device does not take it.
  • A break is a connection that drops and comes back within 10 minutes, as after reloading the connector tab. Every break per day online takes 3 points, up to 30.
  • Stop sharing, stopping in the app, signing out and Ompera's own updates are not breaks.
  • The rating appears after 10 requests that count, or after a day online.