# Dedicated deployments

[RightNow AI](https://www.rightnowai.co/) / [Open models](https://www.rightnowai.co/docs/open-models) / [Dedicated deployments](https://www.rightnowai.co/docs/open-models/dedicated-deployments)

Updated 2026-11-21

A dedicated deployment runs a model on replicas held for your workspace, under a model id of its own. It is for a team that wants to pay for a model by the hour, with the requests on it included.

## How it works

**Illustration of how a deployment works.**

```text
   the dashboard, under Capacity
   │
┌──▼── choose ─────────────────────────┐
│  │  a model, a name and how many     │
│  │  replicas                         │
├──▼── review ─────────────────────────┤
│  │  the price of the day per hour,   │
│  │  what is charged now and what it  │
│  │  covers                           │
├──▼── pay ────────────────────────────┤
│  │  its first 3 hours                │
├──▼── the start ──────────────────────┤
│  │  it starts by itself              │
│  │  if not live within 60 minutes,   │
│  │  everything charged is returned   │
├──▼── watch ──────────────────────────┤
│  │  Starting, then Live              │
│  │  its page shows its model id      │
└──▼───────────────────────────────────┘
╔══════════════════════════════════════╗
║ the stop                             ║
║ start it again under its name, as a  ║
║ new purchase                         ║
╚══════════════════════════════════════╝
```

A deployment is made in the dashboard, under Capacity. Choose: a model, a name and how many replicas. Review: the price of the day per hour, what is charged now and what it covers. Pay: its first 3 hours. The start: it starts by itself, and if not live within 60 minutes, everything charged is returned. Watch: Starting, then Live, and its page shows its model id. The stop: start it again under its name, as a new purchase.

You deploy in the dashboard, under **Capacity**.

## What it costs

A dedicated deployment has no fixed public price. The dashboard shows the price of the day, per hour, before you pay, and that price holds for the purchase.

The price of the day is refreshed during the day. If it changes while you review it, the purchase is refused and the dashboard shows the new price. A purchase is also refused when the price is stale or nothing is in stock. A refused purchase charges nothing.

Requests on the deployment are included in the hourly price: its tokens are not billed.

Its first 3 hours are charged when you pay, before it starts. Its paid time counts from when it is live. After the first 3 hours, each hour is paid before it begins.

It is paid from your balance, the same balance that pays for tokens.

With **Use shared capacity when full** turned on, requests above your replicas run on shared capacity, at the list rates.

## Starting and stopping

### The start

A start is picked up within 1 minute of the payment, and the deployment starts by itself. While it starts, the dashboard reads **Starting replicas**, **Loading the model** or **Checking answers**.

Requests to its model id are refused until it is live.

If it is not live within 60 minutes, it is taken down and everything charged is returned. The dashboard then reads **Failed**.

A start can fail for one of these reasons:

- Nothing was in stock.
- The limit of replicas was reached.
- A replica could not be launched.
- A replica never came up.
- A replica was taken back.
- Its set-up failed.
- Its address failed.
- Its health check never passed.
- The deployment ended while it was starting.

### States

- **Starting**: Paid, and not live yet.
- **Live**: All replicas are live.
- **Partly live**: Some replicas are live.
- **Down**: No replica is live.
- **Stopping**: It serves until the end of its paid time. Requests already running finish first.
- **Stopped**: Nothing is charged. Its name can be started again.
- **Failed**: It did not start. Everything charged was returned.
- **Ended**: A card payment was disputed or flagged as fraud, the model is no longer offered, or the workspace was deleted.

While a replica is down, **Use shared capacity when down** decides what happens to the requests it would have run:

- On: they use shared capacity at no extra charge.
- Off: they are refused, and the down minutes are returned.

A deployment of 1 replica has no spare. While it is down, that choice applies to every request.

You choose both settings when you deploy, where the model offers them.

### The stop

An owner can stop a deployment in two ways:

- Now. Unused paid time is returned to your balance, except its first 3 hours.
- At the end of its paid time. It serves until then, and an owner can undo it until then.

Money is returned to your balance, never to a card.

A stopped deployment stays listed for 31 days, then leaves the list by itself. Its name can be started again at any time. Starting it again is a new purchase, at the price of that day, with a new first 3 hours.

Its page in the dashboard shows what was charged, what was returned and its newest 20 events.

## When the balance runs low

A deployment is paid ahead: its first 3 hours at the start, then each hour before it begins.

Before you pay, the review shows the time your balance covers it until.

To keep it running, add funds or turn on auto-recharge. With auto-recharge on, your card may be charged for it.

When the balance cannot pay for the next hour, the deployment serves to the end of its paid time and then stops. The dashboard then reads **Stopped for low balance**.

Add funds, then start it again. That is a new purchase, at the price of that day.

## Limits

- **Who deploys**: The account owner, in the dashboard. An API key cannot deploy, stop or start a deployment.
- **What pays**: Only credits you bought pay for a deployment.
- **Models**: The models the dashboard lists for a deployment. A model that is not listed is not offered.
- **Replicas**: As many as the dashboard shows can start now. The number is set at the start and does not change while it runs.
- **Name**: 1 to 32 characters: lowercase letters and digits, in words joined by single hyphens. One running deployment for each name.

## In the API

A deployment has a model id of its own: the base model's id, a colon and the deployment's name.

**request**

```shell
curl https://api.rightnowai.co/v1/chat/completions \
  -H "Authorization: Bearer $RIGHTNOW_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MODEL_ID:support-bot",
    "messages": [{"role": "user", "content": "Summarize this ticket."}],
    "max_tokens": 512
  }'
```

`MODEL_ID` stands for the base model's id, and `support-bot` for the name. The id exists once the deployment exists, and its page in the dashboard shows it.

The address and the keys are the same: every key of the workspace can call it. Requests with the base model's id stay on shared capacity.

`GET /v1/models` lists the deployment's id to the keys of its own workspace only.

The answers of `POST /v1/chat/completions`, `POST /v1/responses` and `POST /v1/messages` carry the header `X-RightNow-Capacity`. It says where the request ran:

- **`dedicated`**: The deployment ran the request.
- **`shared; fallback=1`**: The request ran on shared capacity while the deployment was not ready.

The API refuses a call to a deployment with one of these codes:

**Refusals of a deployment**

| Code | Status | When |
| --- | --- | --- |
| `dedicated_deployment_not_ready` | 503 | It is not ready, and **Use shared capacity when down** is off. The answer carries `Retry-After`. |
| `dedicated_deployment_stopped` | 409 | It is stopped, or past its paid time. |
| `dedicated_capacity_full` | 429 | Its requests in flight are at its size, and **Use shared capacity when full** is off. |

A model id of another workspace, or one that does not exist, answers 404.

## Related

[### Limits How to wait on a 429 and send the same request again.](https://www.rightnowai.co/docs/open-models/rate-limits) [### Pricing and credits The list rates of tokens, the one balance, and how to add funds.](https://www.rightnowai.co/docs/open-models/pricing) [### Batch A file of requests at half of list price.](https://www.rightnowai.co/docs/open-models/batch)

## Next

[Models](https://www.rightnowai.co/docs/open-models/models)
