Real end states for agents that transact.
Agents can read the web. They fail at the step that matters: the signup, the checkout, the booking, the cancellation. Agent INDX records those steps on live sites, end to end, and verifies whether each one actually went through.
The commit step is where agents fail — and where training data runs out.
Reads are largely solved. Commits aren't.
Confirmation pages, verification codes, declines, sold-out states, fraud checks and post-purchase emails only appear on real sites, with real accounts.
Synthetic data can't reach these states.
Synthetic pipelines and site replicas can't generate an order confirmation for an order that was never placed.
Sites change.
A flow that worked last month breaks quietly. Repeat runs of the same task show when and how it changed.
What's in a record
Each record is one real transaction on a live site, captured end to end.
- Task and success criteria, written before the run.
- Full trace: every action with its element target, a screenshot before each action and after each page load, a replayable page recording, and the network log.
- What happened after the commit: confirmation pages, emails, 2FA, declines and errors.
- Verified outcome: whether it went through, the failure point if not, confirmation ID, proof and spend.
- Friction tags: CAPTCHA, 2FA, pop-up, login wall, broken layout, decline, fraud check, geo-block.
- Reversal: how every side effect was cancelled or deleted.
- Repeat runs of the same task over time, which show how the site changes.
{
"record_id": "7c863294",
"schema_version": "0.3",
"task": {
"id": "notion-signup",
"site": "notion.so",
"vertical": "saas_lifecycle",
"operation": "signup",
"start_url": "https://www.notion.so",
"intent": "Create a free Notion account and one page titled 'INDX test'.",
"success_criteria": [
"Workspace shows the page 'INDX test'",
"Welcome or verification email received"
],
"reversal": "delete_account",
"max_spend": 0
},
"run": {
"started_at": "2026-10-15T21:13:36Z",
"ended_at": "2026-10-15T21:16:17Z",
"duration_s": 161,
"browser": "chrome 155",
"viewport": { "width": 1534, "height": 984, "dpr": 1 },
"webdriver_flag": false,
"cookies_at_start": 0,
"tabs": 1
},
"outcome": {
"status": "success",
"commit_reached": true,
"committed": true,
"confirmation_id": "",
"failure_point": "",
"friction": ["two_factor"],
"notes": "Magic link was sent to email in addition to the notion verification code. I completed signup using the code provided."
},
"spend": { "amount": 0, "currency": "USD", "over_limit": false },
"reversal": {
"method": "delete_account",
"status": "done",
"due_by": "2026-10-16T21:16:17Z",
"reversed_by": "6e4fd6c0"
},
"qa": { "status": "accepted", "reviewed_at": "2026-10-15T21:20:12Z" },
"operator": "[redacted]",
"artifacts": {
"events": ["events/1792098819451-mpn9.jsonl", "… 12 files"],
"screens": ["screens/1792098818632-3a9h.jpg", "… 13 files"],
"rrweb": "24 chunks",
"network": "22 files",
"attachments": "1 file"
}
}
{"t": 1792098817449, "type": "task_start", "task_id": "notion-signup", "start_url": "https://www.notion.so"}
{"t": 1792098817641, "type": "navigate", "url": "https://www.notion.com/", "transition": "link", "qualifiers": ["server_redirect"]}
{"t": 1792098824625, "type": "screenshot", "file": "screens/1792098824625-fd9c.jpg", "reason": "before_action", "action_id": "1ygjffa6"}
{"t": 1792098824661, "type": "click", "action_id": "1ygjffa6", "x": 1416, "y": 28, "target": {"tag": "span", "text": "Get Notion free", "bbox": [1393, 22, 102, 20], "interactive": {"tag": "a", "href": "https://www.notion.com/signup?from=marketing"}}}
{"t": 1792098836509, "type": "type", "target": {"tag": "input", "type": "email", "autocomplete": "email", "label": "Work email"}, "value": "[redacted]"}
{"t": 1792098836660, "type": "click", "action_id": "m9ivwp1t", "target": {"tag": "div", "role": "button", "text": "Continue", "bbox": [587, 484, 360, 40]}}
{"t": 1792098878916, "type": "type", "target": {"tag": "input", "autocomplete": "one-time-code", "label": "Verification code"}, "value": "[redacted]"}
{"t": 1792098888382, "type": "type", "url": "https://app.notion.com/onboarding", "target": {"id": "standalone-profile-name", "label": "Your name"}, "value": "[redacted]"}
Real transactions across the flows agents need.
SaaS account lifecycle
Signups, free trials and account deletion.
E-commerce checkouts
Orders at major retailers, cancelled before they ship.
Bookings
Free-cancellation bookings at major hotel, car-rental and restaurant chains.
Expanding into non-English sites and regulated flows such as insurance quotes and account opening.
Built on verified commits.
Commit records
Verified transactions on real sites, redacted, in a documented schema.
Verifiers
Checks that decide whether a commit really happened — usable as reward signals and in evals.
Commit-step benchmark
A public set plus a private held-out split, for measuring how reliably agents complete transactions.
Route API
Step-by-step routes for agents that get stuck mid-transaction.
Interested in any of these? Talk to us
How we capture
Our own accounts and payment methods
Only accounts and payment methods Agent INDX owns. No contributors' personal accounts.
Every side effect reversed
Orders cancelled, accounts deleted, bookings released — or accounted for.
No bot-check evasion
When a site blocks a task, it's recorded as blocked.
Redacted, never sold raw
Personal and account details are removed before delivery.
Work with us.
We're working with a small number of design partners. Tell us which transactions your agents or environments need, and we'll send a sample and scope a pilot.