# Developer console

The [developer console](https://platform.samreshuuu.ru) is for calling models from your own code through the [Model API](/docs/model-api) or connecting the agent to your code and AI apps: it holds all of the organization's API keys, what they spent, every call and the balance the calls are paid from. The agent and chats do not need it; they work in the app.

## Sign-in and organization

The console opens at its own address and signs in through the app: if you are signed in to the app, it opens without an emailed code and without asking a second factor you already confirmed in the app; otherwise the app's sign-in opens first. Signing out of the app also signs out the console opened through it; signing out of the console leaves the app signed in. From the app it opens with **Console** in the account menu: at the bottom of the sidebar, or by tapping your avatar on a phone; app settings have no console entry. The organization is part of the page address, and the organization switcher sits at the top of the left column. The console does not change the organization you work in inside the app, so you can hand a link to a section to a colleague from the same organization. At the bottom of the left column is your account: its menu has **Account settings** (the Account section of the app's settings: name, photo, language), **Open Samreshuuu**, **API docs**, the console's **Theme** and **Log out**.

## Sections

| Section | What is there |
| --- | --- |
| **Overview** | The console's first screen. The period is at the top right: today by the hour, a week or a month by the day. Tiles: the balance with a **Top up** link, what was spent and saved on cache, how many requests and how many failed, how many tokens and what share was read from cache; under each number, a line over the period and the change against the previous period of the same length (a month has none: the log keeps today and the 30 days before). Below, spend as one line and the split by model and by key: requests, cache share, cache savings and spend in roubles. A dash in the cache share means the host did not report what was read from cache. Until the organization makes its first call, the **First request** block stands above the numbers: create a key, top up the balance and send a request, with a sample in curl, Python and Node.js. |
| **Playground** | A model request without a key or code: the model, system instructions and messages on the left, the answer streaming on the right. Model settings hold temperature, max output tokens, reasoning effort for models that reason, the response format (text, JSON, JSON schema) and extra parameters. The **Code** view shows the same request in curl, Python, Node.js, the Anthropic SDK and for Claude Code. Every run is paid from the organization balance and lands in the call log; under the answer stand its cost and an **In the log** link. **Add to messages** puts the answer in as an assistant message to continue the conversation. **History** keeps the latest runs in this browser only, for 30 days, and is erased on sign-out. |
| **API keys** | All of the organization's API keys: state, spend and spend limit, last call. **Create key** issues a Model API key at once, without an end date or a spend limit, named by its date. The arrow beside it issues another kind: Full access, AI apps only and View only issue an agent key for the [Agent API](/docs/openai-compatible), connectors and the [MCP server](/docs/mcp), and Key management issues a management key, which issues and revokes Model API keys through the API. The key is shown once. Renaming a key, changing its spend limit and revoking it are in the row menu. |
| **Call log** | Every key call of today and the 30 days before, newest first, refused and failed ones included: time, model, host, key, outcome, tokens, cost, cache savings, duration, time to first token and speed. The outcome says how the call ended, not whether it was paid: **Succeeded**, **Canceled** (the client disconnected; what was generated before that is paid) or **Failed**; the exact end word is on the card. An answer that ended in our error is not charged and reads "not charged". Filter by call ID, key, model, host, outcome and period; a call ID search covers the whole log. A call opens as a card on the right, with arrows to its neighbours. **Export CSV** downloads the log under the same filters. Request and answer bodies are not stored. The Integrations tab lists connector calls made with a key. |
| **Balance** | The same organization balance as in the app: top-up, auto top-up, the monthly spend cap, the history, acts and refunding what is left. More in [Payments, invoices and documents](/help/billing). |
| **Models** | Every model a key calls, priced per 1M tokens in the host's currency, already including our 20%: input, cache read and output. A model's name opens its page on the site with each host's price. |
| **API docs** | Opens the Model API documentation on the site. |

## Who sees what

Sections a role does not open are not shown in the left column.

- **Overview**, **API keys** and **Models** are open to any member of the organization. Creating and revoking keys is for those who manage keys: the owner and admins.
- The **Playground**, spend on the **Overview** and the **Call log** are open to those who manage keys, the balance on the **Overview** to whoever manages billing.
- **Balance** is open to whoever manages the organization's billing.

If a section is missing, ask your organization admin.

## What a call costs

A call costs what the model's host that answered it charges, plus our 20%, and is charged to the organization balance in roubles, a dollar price at the official Central Bank of Russia rate of the day of the call. No plan is needed for that. There is no top-up fee. A call that failed (refused or broken) is not charged. More in [Payments, invoices and documents](/help/billing).
