An agent is its instructions, one model and the live data it may use, kept together. Run it from your console, have it run on a schedule and email you the answer, or publish it at yourname.oraicle.app. Every run is paid from the same wallet.
Each run is paid from the same balance, at the listed prices.
Save an agent — a name, instructions, one model and the live data it may use — and run it from your console, on a schedule or from your own code. Describe the job in a sentence, or start from one of the five templates below.
Get started →Open console →Point OpenClaw, or any agent tool that takes a base URL and a key, at our API. We don't run your agent: we serve the model it calls, from 25+ open models, paid per token from the same wallet.
How to set it up →Put an agent on its own page at yourname.oraicle.app, or on your own site with one line of HTML. Anyone with the link can ask it, and it stops at the daily cap you set.
Get started →Open console →Say in a sentence what it should do, or start from a template. You see what it set up before anything is saved.
"Each weekday at 8, check NVDA and BTC and say what moved them."
Switch on only the sources it needs, and set what it may spend in a day. It stops at the cap, never over it.
stocks · crypto · news · cap $1.00 a day
Run it from your console, have it run every day or every weekday and email you the answer, or put it on its own page.
markets-desk.oraicle.app
The console offers these five as starting points. Pick one, change what you like and save it.
Answer a question with facts and where each one came from.
A dated summary of what happened on the topics you name.
Draft a reply to a customer message, in your tone, for you to send.
Prices and moves for the tickers and coins you name.
Turn a company or person name into a short profile with sources.
Every agent has a cap, every run a price, and nothing runs that you did not set up.
$1.00 a day unless you change it, $20.00 at most. It's a hard stop, across the console, the schedule and the page.
Each run is paid from your wallet at the listed prices. There is no subscription for agents.
Your console keeps each run with its answer and its cost. A page run keeps its cost, never the visitor's words.
Stop a run while it works, and switch a schedule or a page off from the console.
Publish an agent and it gets a page of its own: your title, your greeting, your example questions.
Each run is paid from the same wallet, at the same per-token and per-lookup prices as the chat, with no subscription for agents. A published page can use your account's one free site, or go on a hosting plan.
The tokens it reads and writes, at the model's price per million, plus each live-data lookup it makes, at that source's price. A research brief — about 3,000 tokens in and 800 out on GPT-OSS 120B, with one web and one news lookup — costs about $0.007 at the listed prices.
It doesn't remember earlier runs, and you can't reach it in Telegram yet; both are planned. It never runs code or installs add-ons, and it uses only the live data you switch on for it.
Web search and news are shared by every account, so each account can make 10 web searches and 20 news searches a day (UTC), across its agents, the chat and the API. Past that, a run goes on without the search and says so. The other live data has no daily limit.
One thing: ask that one agent a question. The key sits in the page's source, where anyone can read it, so it can't change the agent, pick another model, see your balance or reach anything else in your account. The daily cap is what protects your wallet, and you can rotate the key or unpublish the page from the console.
The page says it has reached today's limit and answers nothing more until the next day (UTC); a question refused at the cap costs nothing. The cap is $1.00 a day unless you change it, $20.00 at most, and publishing needs a balance at least as large as the cap.
Yes, today. Point OpenClaw, or any agent tool that takes a base URL and a key, at our API with a key from the console. We don't run your agent; we serve the model it calls, and each call is paid per token from your wallet. Use your own agent with Oraicle →
Yes. Send a chat completion to our API with "agent:" and its name as the model, for example "agent:Markets check", using a key from the console. It runs with its own model, instructions and live data, is paid like a run from the console, and stops at its daily cap. Call an agent from your code →
No. Describe the job in plain words or start from a template, and see what it set up before anything is saved. Bringing your own agent tool means editing its settings file; the docs page shows every line.
The same 25+ open models as the chat — DeepSeek, Qwen, Kimi, GLM, Llama, gpt-oss and more. The console offers live data only on the models that can use it.