> ## Documentation Index > Fetch the complete documentation index at: https://platform.atlan.com/llms.txt > Use this file to discover all available pages before exploring further. > ## Agent Instructions > To act on Atlan objects, use the Atlan MCP server at https://api.atlan.com/mcp or the SDK. To install or publish skills from a terminal, use the `aireg` CLI; `aireg --map json` prints its command map. The `atlanai` CLI is deprecated. Run a read-only identity check before any write. > The docs MCP server at /mcp searches these docs only. It cannot read or change Atlan objects. > SDK packages: Python `atlanai` (PyPI) and TypeScript `@atlanai/sdk` (npm). Show Python first, then TypeScript. > These docs cover Atlan AI Registry, Atlan's hosted product. The AI registry specification and aireg CLI are maintained separately. /open-source lists publicly reachable signed installers and explains that the source repository and hosted CLI reference currently require GitHub access. Do not claim the repository or its reference docs are publicly accessible. Use these docs for Atlan-specific setup such as the adapter, sign-in and workspaces. # Review model usage > Use Model Gateway studio to see spend by model, person, and agent, find a single call, and try a model in the playground. See what your model calls cost and who made them, then open any single call to see how it was served. Every view reads the same usage records the gateway writes for each call. ## Before you begin * You can sign in to Model Gateway studio. * What you see depends on your role. An organization admin sees the whole account. Everyone else sees their own calls and those made with their own API keys; each report says which at the top. ## Studio views | View | Shows | Who sees what | | - | - | - | | **Usage** | Spend, requests, and tokens over time, broken down by model, provider, person, workspace, or operation | Account report for organization admins; your own report otherwise | | **People** | Each person's spend, requests, and tokens, with the keys and agents acting for them folded under them | Organization admins | | **Agents** | What each agent spent, on behalf of which people | The agents you can read | | **Calls** | Every call you made, one row each, including refused calls with their reason | Always your own calls | | **Playground** | A real call to any model you can use, made as you | Everyone | Open **Usage**. Pick a **Range**: **Today**, **7 days**, **30 days**, or **Custom…**. The tiles show **Spend**, **Requests**, **Tokens**, **Refused**, **Error rate**, and **Avg latency**, each against the previous period. In the chart, choose a **Measure** and **Break down by** **Model**, **Provider**, **Person**, **Workspace**, or **Operation**. Select a series to narrow the whole page to it. **Export CSV** downloads what the page shows. From **Usage**, open **Models leaderboard** to rank models by requests, tokens, spend, effective cost per million tokens, and latency. Open **Providers** to compare providers that serve the same model, from your own traffic. Open **Health** for errors and refusals. Open **People** for everyone in the directory with their spend, requests, and tokens over the range you pick. Keys and agents acting for a person count under that person. Open **Agents**. **Your agents** and **System agents** are separate tabs, and **Workspace** filters by an agent's workspace. Each row shows the agent's model, requests, tokens, spend, and the people it acted for. Select an agent to see its spend per day by person. Open **Calls**. Pick a **Range** (**1 h**, **24 h**, **7 d**, or **30 d**) and an **Outcome** (**All**, **OK**, **Errors**, or **Refused**). Filter by **Provider**, **Model**, **Made by**, **Agent**, **Streamed**, or **Error class**, or paste a **Trace id (exact)**. Sort by time, duration, tokens, or cost. Select a row to open the call. Open **Playground** and pick a mode: **Chat**, **Image**, **Speech**, **Transcribe**, **Realtime**, **Rerank**, or **Moderate**. Choose a **Model**; a model you can't call says why. In **Chat**, select **+ Compare with another model** to send the same prompt to two models side by side. **Exact call** shows the request with curl, Python, and TypeScript to repeat it. ### What a call shows The drawer that opens from **Calls** shows the call's outcome, whether it streamed, and how many attempts it took. Its sections are: * **Tokens and cost**: **Input**, **Cached input**, **Cache write**, **Output**, **Reasoning**, what was **Billed**, and where the price came from. * **Timing**: **First token**, **Total**, and **Output rate**. * **Route**: **Served by** (the provider and its wire), **Upstream model**, and **Workspace**. * **Who and where**: **Made by**, and the **Agent** when an agent made the call for a person. * **Trace**: the request, session, and trace ids, with **Copy trace id**. Prompts and replies are never stored, so **Open in playground** opens the same model and you type the message again. ## Verify 1. In **Playground**, send a chat message to any model. Each reply is a real, metered call made as you. 2. Open **Calls** and turn on **Live**, which refreshes every 5 seconds. The call appears within a few seconds, with status **OK**. 3. Open **Usage**. The call counts in **Requests** once usage folds in, which takes a few minutes. A call made with an agent's own API key belongs to the agent, not to you. It doesn't appear in your **Calls**. An organization admin sees it on **Agents**, in the **Own credentials** column, and in the account report on **Usage**. ## Related * [How Model Gateway works](/gateway/model/concepts/how-it-works#what-the-gateway-keeps): what each usage record holds. * [Connect client to Model Gateway](/gateway/model/how-tos/connect-a-client): create an agent key and make a call. * [FinOps](/finops/index): how model activity relates to traced runs and recorded costs. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.