> For the full page index, fetch https://echonova.ai/llms.txt # What a turn costs What a voice and a text conversation are charged for, how AI usage becomes credits, and a worked voice turn and text turn at today's rates. A turn is one exchange: the customer says something, the agent answers. A voice conversation is charged for the time it is connected and for the AI model work in each turn; a text conversation is charged for the AI model work only. This page explains both parts, how they become credits, and where to read the number afterwards. ## What is charged | Part | When | Price | | -------------- | ------------------------------------------------------------------------------------------------------------------------------------------ | ------------------------------------------------------------------------------------------- | | Connected time | Voice conversations: web calls, phone calls and WhatsApp calls. From the moment the call connects to hang-up, silence included, per second | 2,500 credits per minute | | AI model usage | Every turn, voice and text | The provider's price for the tokens the model read and wrote, plus 10%, turned into credits | | Call summary | Once, when the conversation ends and its summary is written | The same AI model price, on the same conversation | The speech in a voice call, listening and speaking, is covered by the 2,500 credits a minute. It is not charged again per second of audio or per character. ## How a turn is charged Each turn charges the connected time since the previous turn and the turn's AI usage together, in one charge. When the call ends, the time after the last turn is charged. AI usage is priced in dollars first: the tokens the model read, the cached tokens and the tokens it wrote, each at the model's own price, plus 10%. The dollars are then turned into credits at the rate of the credits that pay for them; the rates are on [Credits](/docs/developers/credits#how-ai-usage-becomes-credits). The model picker shows each model's provider price per 1M tokens and an estimate of a call minute on your plan; see [Choose a model and reasoning effort](/docs/agents/model-and-reasoning). ## A voice turn, worked A call on Starter monthly is connected for 2 minutes 15 seconds. The connected time costs 135 × 2,500 ÷ 60 = 5,625 credits, whatever was said. One of its turns reads 1,200 tokens and writes 34; if those tokens cost $0.004 at the provider's price, the turn is billed $0.0044, which is about 139 credits on Starter monthly. The prompt is the biggest part of the AI usage, and it is the one you control. Long instructions, a high **Top documents** setting on retrieval and a long conversation history are all read again on every single turn. ## A text turn, worked The same question on a text channel. There is no connected time, so only the AI usage is charged: about 139 credits for the same tokens. A written reply is often a little longer than a spoken one, so its output tokens can be higher. ## Get it right The prompt is the biggest line on a voice turn and nearly the whole of a text turn, and it is read again on every turn. Keep what the model reads short. **Recommended** - Instructions that fit on one screen, with reference material in knowledge and **Top documents** at 3 to 5. - A reply style that answers in one or two sentences on a call. **Less effective** - The whole handbook pasted into the instructions, so every turn reads it again. - A voice agent that reads out long lists. Every minute of the call is charged, so long answers lengthen it. ## What tool calls cost Nothing directly. A tool call is counted on the conversation and shown on it, and it carries no price of its own. It is not free, though. A tool call adds a round trip through the model: the model is asked once to decide on the call, the result is fed back, and the model is asked again to answer with it. Both of those read the prompt again, and the turn's token counts cover both, so a turn's token count and its tool count move together. ## Where to see it | Where | What it shows | | ------------------------------------------------ | ------------------------------------------------------------------------------------------------------ | | A conversation's detail panel, **Total credits** | Every credit the conversation was charged, added up across its turns | | A conversation's **LLM cost** chart | The model's share of those credits over the life of the conversation | | A conversation's **Tools triggered** | How many tool calls ran, with the names of the tools | | The agent's Overview tab, **Credits used** | Credits charged across all of this agent's conversations since the first of the current calendar month | The model's share is a proportion of the same credits, not a second charge, so it is always less than the total rather than being added to it. On a text conversation it is nearly all of it. Every turn adds to the conversation's running usage as it happens, so the numbers on a live conversation climb while you watch them rather than appearing at the end. ## Playground sessions cost the same A session started from the **Test** button is metered the same way, at the same rates, against the same balance. There is no test mode, no sandbox rate and no free allowance. The only difference is who is talking. See [Test your agent](/docs/agents/test-your-agent). ## What is not charged - An agent that is sitting idle. Configuration costs nothing. - Editing an agent, adding a voice, changing the model or tuning retrieval. - A tool call, on its own. - A text conversation that fails before its first turn. A voice call is charged for the time it was connected, even when it ends before the first turn. Indexing knowledge, crawling a website, building an agent with AI and setting up a database tool use AI models. They are charged for that AI usage, the same way as a turn. ## When the balance runs out A call or chat can start only while the balance is at least 1 credit, and it is checked again on every turn. A turn that cannot be charged is the last one: the conversation ends, its status reads `out_of_credits` and, for a session you started yourself, a notification reads **Voice session ended: out of credits**. The turns already charged stay charged. Nothing is refunded and nothing is lost from the transcript. On a trial, AI usage stops once its provider cost reaches $1: the conversation ends and the portal shows **Your trial AI allowance is used up. Upgrade to keep using AI.** ## FAQ ### Is a tool call charged? Not on its own. It adds a model round trip to the turn, so the turn's prompt and completion tokens are higher than they would be without it. ### Why does a voice turn cost more than a text turn? A voice conversation also pays 2,500 credits for every minute it is connected. A text conversation pays for the AI usage only. ### Do I pay for the model's reasoning? Yes. Completion tokens include any reasoning the model did before answering. See [Choose a model and reasoning effort](/docs/agents/model-and-reasoning). ### Do rates change? The per-minute price of a call is set on the rate card and can change; AI usage follows the provider's price. A turn is priced at the rates in force when it runs, and the credits recorded on the conversation do not change afterwards. ### Are playground sessions cheaper? No. They are charged like a customer conversation. ## Related - [Test your agent](/docs/agents/test-your-agent): a test session, charged the same way - [Agent status](/docs/agents/agent-status): what keeps charging after an agent is archived - [Working hours](/docs/agents/working-hours): conversations outside the window are charged too - [Choose a model and reasoning effort](/docs/agents/model-and-reasoning): the model's share of the charge - [Voices and language](/docs/agents/voice-and-language): the voice a call speaks with - [Credits](/docs/developers/credits): the conversion and the rates - [What is an agent](/docs/agents/what-is-an-agent): what a turn belongs to