What will your AI agent actually cost?
Every Diali plan includes a monthly balance of AI credits. Describe how you would use an agent and see what that month costs, and which plan covers it.
Free tool
Loading live prices
How Diali AI credits work
A credit is a thousandth of a US dollar, and every plan comes with a monthly balance of them. When your agent thinks, it spends credits at the model's published rate: so much per million tokens read, so much per million written. A few tools, such as managed web search and image or video generation, are priced per use instead, and unused credits do not roll over.
That is the whole mechanism. There is no per-message fee, no seat, and no separate AI subscription to hold. It also means a quiet month costs you nothing extra, and a heavy one is visible in the same units the estimate above uses.
You can also bring your own key from any of the supported providers and pay them directly instead. The agent runs the same either way. Credits exist so that nothing else has to be signed up for before an agent can think, not to stop you using an account you already have.
Why OpenClaw API costs run away
OpenClaw bills get out of hand in the same three places almost every time. The context snowballs: an agent re-reads its instructions, its tools and the recent history on every turn, so a long conversation sends far more text than the visible messages suggest. The heartbeat keeps thinking while nothing is happening, which is how an idle agent still spends money every day. And the expensive model ends up on jobs that never needed it.
“I left the setup as it was, and now I’m being charged $5 daily for nothing.”
“Burned over 40 million tokens so far, and I have no idea how.”
The calculator above prices all three honestly. Repeat context is the cached share, billed at a fraction of the rate. Messages a day is the volume, heartbeat included. And the model picker shows what the same month costs on a cheaper model, which is usually the biggest lever of all.
Quotes are from public posts, lightly trimmed.
Questions people ask
The prices are live from our catalog, so the rates are exact. The usage is your guess, and that is the part that moves: a month of real conversation rarely matches a tidy average. Treat it as the right order of magnitude rather than a bill.
Providers charge a fraction of the normal rate for input they have already seen recently. An assistant with memory re-reads its instructions and recent history on every turn, so a large share of a long-running agent's input is cached. Ignoring it would overstate the cost of exactly the agents people keep.
Your agent pauses its thinking rather than running up a bill you did not agree to. You can top up at any time, or connect your own provider key and carry on. Nothing is deleted and nothing is charged automatically.
Stop estimating, start running one
Every plan includes credits, every channel and every connector. Build an agent and watch what it actually uses.