Introduction
OpenIntents lets your agents buy anything. Your agent says what it wants in one sentence, such as "buy a flat white from the café on 5th", "book an Uber to SFO at 7am" or "reorder my coffee beans on Amazon", and it gets done: found, checked out and paid, with a structured receipt at the end. Your agent never has to drive a browser itself.
There are no merchant integrations to build and no SDK per store. If it has a checkout on the web, it's an intent.
How an intent gets done
- You call the API with an intent in natural language, from your code (REST API), an AI agent (MCP) or the CLI.
- The API does the agentic work. Behind the endpoint, OpenHuman drives a real browser: find the merchant, pick the item, fill in details, and check out, the way a person would.
- The call resolves to a receipt: merchant, items, total, fee and confirmation number, as JSON. Optionally, a webhook fires when it's done.
json
{
"id": "int_7f3a2c",
"status": "completed",
"intent": "flat white, oat milk, pickup 8:45",
"result": {
"merchant": "Blue Bottle Coffee",
"total": 550,
"fee": 28,
"currency": "usd",
"confirmation": "BB-48213"
}
}Free compute, pay per transaction
Running an agent in a browser takes inference, a machine and bandwidth. We cover all of it. You pay a fee only when a purchase actually goes through.
| What | Cost |
|---|---|
| Inference | Free |
| Browser sessions | Free |
| A machine per intent, compute and bandwidth | Free |
| Each completed transaction | 5% of the total |
Failed or cancelled intents cost nothing. See Pricing.
Where next
- Quickstart: send your first intent in under a minute.
- Intents: the object, statuses and lifecycle.
- Balance & payments: paying from a balance or a payment link.
- SDKs for TypeScript, Python and Rust, the CLI and the OpenAPI spec are open source: github.com/tinyhumansai/openintents.
- Building with an LLM? Point it at
/llms.txt; every page is also available as raw Markdown.