Get started
Quickstart
Put a running application behind Novie Control in about ten minutes, using a Run workload and its Application Key.
This guide puts a running application (a product feature, an internal service, an agent in production) behind Novie. When you finish, every model call your app makes is budgeted, checked against policy, and reported against one accountable workload. The only things you change in your code are the base URL and the API key.
Before you start
- A Novie Control account at control.novamindlabs.ai with permission to create workloads in your unit.
- A unit (department or business unit) in Structure that has budget it has not yet given to workloads. If it has none, ask a finance or organisation admin to allocate some in Budgets.
- An application that calls an OpenAI-compatible or Anthropic-compatible API and lets you set the base URL and API key.
1. Create a Run workload
Open Workloads → New workload, or select your unit in Structure and choose New workload.
- Unit and name: choose the unit that pays for it and give the app a recognisable name, for example Checkout API.
- Class: choose Run, for a running app that spends through one Application Key.
- Financial tag: choose CapEx or OpEx. It locks once the workload has spent money.
- Application Key expiry: choose whether the key expires at the end of the fiscal year or never expires.
- Owner and members: the owner is accountable for the spend. Members can see and manage the workload.
- Links (optional): the repository and the live URL where the app runs, so reviewers know what they are approving.
- Budget: enter an amount above zero, up to what the unit still has available. Run workloads always need a budget.
Choose Create. The budget is set aside from the unit immediately.
2. Create the Application Key
Right after the workload is created, Novie moves to its second step, Application Key. Choose Create key and copy the secret: it is shown once. Store it in your app's secret manager, not in source code.
If you skip this step, create the key later from Keys → Create Application Key and pick the Run workload. A Run workload holds one active Application Key at a time.
3. Wait for approval, if policy requires it
Your organisation's policies decide whether the new workload is approved automatically or waits for a reviewer. If it waits, it appears in Approvals, and the confirmation tells you so. The app can spend once the workload is approved; until then calls are refused and nothing is charged.
4. Point your app at Novie
Replace your provider's base URL and API key with Novie's. No extra headers are needed: the gateway finds the workload from the Application Key.
| Setting | Value |
|---|---|
| Base URL | https://gateway.api.novamindlabs.ai/v1 |
| API key | Your Application Key |
curl https://gateway.api.novamindlabs.ai/v1/chat/completions \
-H "Authorization: Bearer $NOVIE_APPLICATION_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "<model id from /v1/models>",
"messages": [{ "role": "user", "content": "Hello from Novie" }]
}'import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://gateway.api.novamindlabs.ai/v1",
apiKey: process.env.NOVIE_APPLICATION_KEY,
});Call GET /v1/models with the same key to list the models this workload may use. Besides Chat Completions, the gateway accepts the OpenAI Responses and Embeddings APIs and the Anthropic Messages API at the same base URL.
5. Watch it in Control
- Workloads: the workload's balance goes down as calls settle.
- Usage: each request, its model, tokens and cost.
- Overview: spend rolled up by unit.
Keep it running
- Top up automatically: give the workload a daily, weekly or monthly scheduled top-up so it never runs dry. See Budgets and funding.
- Rotate the key: on Keys, rotating replaces the secret and keeps the key's settings. Deploy the new secret, and the old one stops working.
- Restrict models: attach a policy to the workload or its unit to control which models it may use. See Policies.
- When something fails: see the error codes in Models and the AI gateway. A
402means the workload is out of budget.
Building rather than running?
If you are using AI while you build (in Claude Code, Codex, scripts or notebooks), use a Build workload instead. Create a personal Novie Key on Keys, and you spend from the Build workload you select. See Workloads and the Novie daemon.