Product

What happens when an AI agent logs in.

OneHuman spots the agent, applies your rule to every action, and signs each decision on your own server.

Try it live →Book a demo

Every rule assumes a person clicked. Four questions now have no answer.

Who agreed?An export or a transfer was clicked. Nobody can show a person meant it.Consent

Where did the data go?Private data leaves through your front end into a model you did not pick.Data

Who pays for the mistake?An injected instruction moves money. The dispute lands on you.Liability

What will the auditor see?A valid session and a click. Nothing says who acted.Audit

One session, step by step. A person, then their AI agent.

A customer signs in

Their clicks move like a hand. The action returns the balance.

An AI agent attaches

Seen in 0.3 s. What is on screen is hidden before the agent reads it.

The action decides

Balance hidden. The export waits for the person. Every decision signed.

The person approves

Only their passkey approves an action or takes the session back.

bank.example.com

Session protected

Balance$4,939.10

NameAda Lindqvist

IBANGB29 NWBK 6016 1331 9268 19

Phone+1 (415) 555-0142

BalanceStatement

✦Assistant

What’s my balance?

Reading the page…

Verify it’s youTouch ID · passkey

decision balance.read → allow · actor=human_like · HUMAN_KINEMATICS

How it tells a hand from a program. Three pointers, live.

pointer check · live

HumanCurved, trembling, slows onto the target. Agent driverJumps to the target, clicks in 1–4 ms. Humanised botA smooth curve, too clean for a hand.

Spot, decide, prove. Then hand back to the person.

Spot.

See every AI agent in your customers' sessions the moment it joins. Start in watch-only mode.

Decide.

One rule per action: allow it, hide private fields, ask the owner, or never share.

Prove.

Every decision is signed and chained on your server. An auditor checks it offline.

Hand back.

Only the owner's passkey approves a risky action. The agent cannot do that step.

The numbers behind this: How we measured → · Agent scorecard →

People now use their bank, their CRM and their work tools through an AI agent.Most apps can only block it or trust it.OneHuman lets it in on your terms.

Try it against a real agent. Three demo apps, your rules.

Loading…

Questions

More on the home page, or write to hello@onehuman.ai.

How does it tell an AI agent from a person?

Two ways. Agent tools leave traces in the page when they attach, so the session is marked before the agent's first action. And every click is checked: a hand curves, trembles and slows onto a button, while a program jumps there and releases in a few milliseconds. Agents that sign their requests (Web Bot Auth) are identified by signature.

What do I need to install it?

A Node.js server (22.13 or newer): Express 4 or 5, Connect, a Next.js custom server or plain node:http. Run npx onehumanai init in your project. It finds the routes worth protecting, asks three plain questions and shows every change before it writes it.

Is it open source, and what does it cost?

The browser SDK, the middleware and the CLI are Apache 2.0. The engine's source is public under the Business Source License 1.1, with production use granted. You install it from npm and run it on your own server.

Try it live →Read the docs →

Know who agreed.Book a demo