---
title: Chat
description: Talk to any chat model from the dashboard, tune the system prompt, temperature and output limit, and see the speed and cost of every reply.
---

Open [Chat](/app/playground) from the sidebar's **Playground** group. It talks to chat models through the [Chat Completions API](/docs/models/openai), streams every reply, and shows how fast it was and what it cost. Calls are billed to your balance and show in [Logs](/app/logs), like any [playground](/docs/playground) call.

## Settings {#settings}

The settings sit in a panel on the left. On a phone, the model picker stays at the top and the rest opens with **Options**.

| Setting | What it does |
| --- | --- |
| **Model** | The chat model to talk to. It starts on Claude Sonnet 5.5 when your account has it. Only models that speak Chat Completions are listed; image models are in [Images](/docs/playground/images). |
| **Channel group** | **Auto — best available** lets Lovely Router choose. Pick a group to send the request to it, as the `x-lovely-group` header does from code. See [Channel groups](/docs/models/channel-groups). |
| **System prompt** | Optional instructions sent before the conversation, up to 20,000 characters. |
| **Temperature** | From 0 to 2, in steps of 0.1. Starts at 0.7. |
| **Max output tokens** | Empty uses the model's own limit (**Model default**). Reasoning models spend output tokens on thinking first, so a small cap can cut the answer short or leave it empty. |

Below the settings, **Price** shows what the model costs you per million input and output tokens, with your channel group's multiplier applied.

A model that runs on our own GPUs shows an **Asleep**, **Waking up** or **Warm** badge under the picker. See [Asleep, waking up, warm](/docs/playground#hosted-models).

## Write for me {#write-for-me}

The **System prompt** field has a writing helper in its corner. On an empty field it offers **Write for me**; once there is text it offers **Improve**. Its menu also has **Shorter**, **Write a new one** and the **Writing model**, which is shared by every "Write for me" in the dashboard. The text streams into the field, **Undo** brings your version back, and the cost of the call is shown. Each use is a chat call billed to your balance.

## Send messages {#send-messages}

Type in the box at the bottom and press **Send**, or Enter. Shift+Enter adds a new line. While a reply streams, **Send** becomes **Stop**, which ends the reply where it is.

Until the first token arrives you see **Thinking…** with a seconds counter. A reasoning model's thinking appears in a **Thought process** section above its answer; it starts open and you can collapse it.

Each request sends the whole conversation so far, with the system prompt first, so the model sees the full history. **Clear** in the header starts a new conversation. The conversation lives only in the page: leaving or reloading it starts over.

## Speed and cost {#speed-and-cost}

Under every reply you see:

| Figure | Meaning |
| --- | --- |
| **TTFT** | Time to first token. |
| Total time | From sending to the last token. |
| **tok/s** | Output tokens per second after the first token. |
| **in → out tok** | Input and output tokens the model reported. |
| **≈ $** | The estimated charge, from the tokens and the price shown. |

The exact charge, and the channel that served the call, are in [Logs](/app/logs). If a request fails, the error message appears in place of these figures.

## Get the code {#get-the-code}

**View code** in the header shows a cURL command for `POST /v1/chat/completions` with the current model, temperature, output limit, system prompt and the message you are typing (or `Hello`). It doesn't include earlier turns of the conversation. Replace `$LOVELY_API_KEY` with a key from [API keys](/app/keys) and run it. **Hide code** closes it.

To do the same from your own code, with streaming and usage, see [OpenAI Chat Completions](/docs/models/openai) and [Streaming](/docs/models/streaming).
