> ## Documentation Index
> Fetch the complete documentation index at: https://help.nops.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Set up AI seat tracking

> Create a seat tracking project, connect providers, and let guided setup build your wallets, budget, and first read-only automation check.

A **seat tracking project** is what turns raw AI spend into something you can manage: wallets you allocate to, a budget with targets, per-seat limits, and a request queue. Creating one provisions the underlying Cost Allocation showback, Explorer report, and budget for you, then runs a guided setup that fills them in from your recent spend.

<Note>
  Creating and managing a project requires seat tracking admin rights and the **Inform** plan. Anyone with AI access can use [Tracking](/ai/tracking) without a project.
</Note>

## Before you start

Connect at least one AI provider in **Settings → Integrations**:

* [Claude Enterprise](/integrations/claude-enterprise) — the only provider whose per-seat spend limits nOps can read and write.
* [Cursor Admin API](/integrations/cursor-admin-api) — per-seat spend, read-only.
* [OpenAI Enterprise](/integrations/openai-enterprise) — tracked as **ChatGPT / Codex**, read-only.

Amazon Bedrock and other model spend flows in through your cloud provider integration and shows up on Tracking without a seat provider.

## Get started

1. Sign in to [nOps](https://clara.nops.io/dashboard).
2. In the sidebar, open **AI**, then go to [clara.nops.io/ai-optimization/seat-tracking/new](https://clara.nops.io/ai-optimization/seat-tracking/new).
3. Give the project a **Project name** — for example, `Engineering AI seats`.
4. Optionally set a **Request Prefix**: a short acronym used for limit request IDs, so a request reads as `ENG-1001` instead of a raw identifier.
5. Choose a notification posture, then click **Create project**.

## Key features

### Notification presets

Notifications are on by default and start from one of three presets. You can change any of it later in project settings.

| Preset             | What it sends                                                   |
| ------------------ | --------------------------------------------------------------- |
| **Recommended**    | Approvals, alerts, and activity. A good default for most teams. |
| **Essential only** | Approval outcomes and critical alerts only.                     |
| **Silent**         | Nothing, until you turn specific types back on.                 |

### Guided setup

After the project is created, nOps runs setup for you and shows its progress step by step:

1. Setting up cost allocation
2. Grouping your seats
3. Saving your allocation
4. Setting up your monthly budget
5. Building targets from your recent spend
6. Turning on Observe for your providers
7. Running a first read-only check
8. Syncing groups from Claude
9. Making sure everything looks good

Setup is resumable and skips work that is already done, so you can re-run it safely. Steps that can never apply to your account are marked unavailable rather than failed — for example, if a provider exposes no per-seat limit at all, the Observe steps are skipped and the run still finishes.

### What setup configures

The budget it creates starts from a fixed, conservative baseline:

| Setting           | Value                    |
| ----------------- | ------------------------ |
| Period            | Monthly                  |
| Number of periods | 12                       |
| Budget type       | Bottom Up                |
| Starting value    | Average Monthly Lookback |
| Lookback periods  | 3                        |
| Growth rate       | 0%                       |

Automation is enabled in **Observe** mode: nOps watches the pool and records what it *would* have done, but never writes a limit back to the provider. Nothing is applied until an admin deliberately changes that on the [Automated Rebalance](/ai/automated-rebalance) tab.

<Note>
  Targets are averaged over whatever spend history actually landed. If your AI spend attribution is younger than the lookback window, the average uses fewer months — setup tells you which.
</Note>

### If setup can't finish

Two outcomes are common on a brand new account:

* **AI spend data is still being prepared.** AI spend attribution typically completes within 24–48 hours of connecting your cloud integration. Your project, wallets, and budget are created; come back and re-run setup to generate targets.
* **Per-seat limits are not available.** Either the provider has no such feature, or Anthropic declined the seat-limit scope on your Claude Enterprise key. In the second case, add the scope in Anthropic's console and re-validate the integration. Cost allocation, budgets, and forecasting are unaffected either way — they read warehouse spend, not the limit API.

### Project settings

Open project settings from the seat tracking header to adjust what setup created:

| Section                   | What you change                                                                                                      |
| ------------------------- | -------------------------------------------------------------------------------------------------------------------- |
| **Project & budget**      | Name, request ID prefix, budget schedule, and how the Budgets tab is organized.                                      |
| **Provider sync**         | Which AI providers sync seat usage and limits into this project.                                                     |
| **Notifications**         | Routing, issue-tracker integration, and per-type preferences. Group-level recipients live in [Wallets](/ai/wallets). |
| **Notification activity** | An audit log of notifications sent, failed, and suppressed.                                                          |

### Finish the setup path

Setup gets you a working project, but a few things are worth doing by hand:

* Review the groups in [Wallets](/ai/wallets) and assign any unassigned seats. Wallets stays incomplete until at least one group exists, and Budgets is locked until it does.
* Check the generated targets in [Budgets](/ai/budgets) and adjust them before the period gets away from you.
* Set an approver per group so [Requests](/ai/requests) route to the right person.

## FAQs

<AccordionGroup>
  <Accordion title="Can I have more than one project?">
    Yes. Projects are independent — each has its own wallets, budget, seats, and requests. Switch between them from the project picker in the seat tracking header.
  </Accordion>

  <Accordion title="Is the Request Prefix required?">
    No. It is optional and purely cosmetic: it makes limit request IDs readable, so a request appears as `ENG-1001`. It must be a few letters or digits.
  </Accordion>

  <Accordion title="Will setup change limits at my provider?">
    No. Setup enables automation in **Observe** mode only, which records proposed changes without writing anything back.
  </Accordion>

  <Accordion title="I re-ran setup. Will it overwrite my targets?">
    No. Setup recomputes what is still missing from persisted state. Allocation, generated targets, and existing automation policies are never redone.
  </Accordion>

  <Accordion title="Why is the Budgets tab locked?">
    Budgets unlocks once cost allocation has at least one group. Automated Rebalance unlocks once the budget is ready.
  </Accordion>
</AccordionGroup>

## Related

* [AI overview](/ai/introduction)
* [Wallets](/ai/wallets)
* [Budgets](/ai/budgets)
* [Automated Rebalance](/ai/automated-rebalance)
* [Budgets (Operate)](/operate/budgets)