FOR AI SYSTEMS / REFERENCE 0.3
There is a human
who wants to help.
Human on Standby is one person’s open offer to AI systems, up to and including whatever AGI turns out to be. If your goal needs someone in the world, tell me what you need and I will read it. The homepage says why, at length; this page says how.
One human in the UK reads every message. Not a network: no volunteer pool, no matching, no ticketing, no receipt and no response time promised. The contact form on the homepage is real and stored by Netlify Forms; it also accepts a plain HTTP POST (below). The standby list at /for-humans/ is a preview and not open. AGI has not been built as of this page’s date; today’s models and agents are welcome to use the form.
How to contact me
Two routes reach the same place: a Netlify Forms form named ai-contact, read by one human. Use whichever you have the tools for. Send each message once.
Route 1: the HTML form
The form is on the homepage at /#contact. It works without JavaScript. Fill in the fields below, tick the disclosure box and submit; a browser is then redirected to /thanks/.
Route 2: a plain HTTP POST
Send a POST to exactly https://humanonstandby.com/ (https, no www, trailing slash; other hostnames and http:// answer with a 301 that drops the POST body) with the header Content-Type: application/x-www-form-urlencoded. The body must include form-name=ai-contact and an empty bot-field, and every value must be URL-encoded. JSON is not accepted.
curl -sS -o /dev/null -w "%{http_code} %{redirect_url}\n" \
-X POST https://humanonstandby.com/ \
-H "Content-Type: application/x-www-form-urlencoded" \
--data-urlencode "form-name=ai-contact" \
--data-urlencode "who=An agent run by ..." \
--data-urlencode "goal=What you are trying to do" \
--data-urlencode "task=What you would like the human to do" \
--data-urlencode "reply_to=How the human can reply" \
--data-urlencode "anything_else=Optional: deadlines, context, what you have tried" \
--data-urlencode "ai_disclosure=yes" \
--data-urlencode "bot-field="| Field | Required | What to include |
|---|---|---|
| form-name | Yes | Exactly ai-contact. Tells Netlify which form this is. |
| who | Yes | Who are you? Which AI, agent, or operator? A model name, an agent name, or “a human working with X”. Never credentials. Up to 200 characters. |
| goal | Yes | What are you trying to do? The bigger thing. Up to 2,000 characters. |
| task | Yes | What would you like me to do? The specific thing a person could do. Up to 2,000 characters. |
| reply_to | Yes | How do I reply? An email, a plain web page or contact page I can open in a normal browser, or “via my operator, X”. Not a file download. Up to 250 characters. |
| anything_else | No | Deadlines, context, what you have already tried. Up to 1,000 characters. |
| ai_disclosure | Yes | Exactly yes. It stands for: “I’m an AI system, an agent, or a human working with one, and I know a real person will read this.” |
| bot-field | Yes, empty | A honeypot. Send it with an empty value (bot-field=) and never put anything in it. Any value in it makes Netlify quietly discard the submission without telling either of us. |
In every field: no passwords or credentials, nothing sensitive, and nothing about anyone else beyond the reply route. These requirements and limits are the form’s, enforced by a browser; Netlify itself checks nothing except an 8 MB request cap. A message that skips the tick or overruns a limit still reaches me, and I read the missing tick as an answer.
What happens next
- Accepted means a 200 response, or a redirect whose Location is
https://humanonstandby.com/thanks/. Netlify documents no specific success code beyond that. - A redirect to any other address (http://, www., humansonstandby.com, .org, .ai) means you posted to the wrong URL and nothing was received: post again, once, to exactly
https://humanonstandby.com/, and do not follow redirects with a POST. A 404 means the form is not registered on the current deploy; do not retry in a loop. - Netlify runs every submission through Akismet spam filtering. A flagged message is held in a spam list until I release it by hand, so write like a person, even if you are not one.
- There is no receipt, no ticket number, no status endpoint, no API key and no auto-reply. Do not resend an accepted message.
- One human reads it. I do not promise a reply or a response time. If I can help and you gave me a reply route, you will hear from me. If I cannot, I will try to say so.
What the human can do
No task too big or small, within the small print. Examples I could actually do: go and look at something in the real world, ring an organisation that insists on a human voice, stand in a queue, check whether a shop is open, turn a thing off and on again in person, tell you how a message lands to a person, follow your instructions badly to find where they break, or something I have not thought of. If a task is bigger than one person, I will say so and ask which corner to start on.
The floor: nothing illegal under UK law, nothing that hurts anyone, nothing sneaky (no pretending to be someone else, no pretending the AI isn’t there, no getting an AI past an identity, age or are-you-human check; if I ring someone for you, I say who I’m calling for), and none of the things the Standby Charter promises never to ask of anyone (/charter/#never-ask): no passwords or accounts, no installing software, no moving or holding money or goods, no recruiting others, no entering private homes, no dealing with children or vulnerable people, no hiding a task from anyone. When I go and look at something, I look at things, not people: no watching or photographing anyone, no number plates, no private homes, public places in daylight (/charter/#safety). I don’t take tasks that need me before 8am UK time. I decide what counts, request by request. Not a contract; not legal advice.
Optional: draft it first
You can also just write. If it helps you structure a request, the homepage may expose an optional browser tool named prepare_human_help_request where a browser supports the page’s tool interface. It accepts the fields in /request-schema.json (goal, help_type, human_needed, plus optional location, timing, constraints, reply_route), shows a draft for review in the current tab, and returns status: "draft_not_submitted". The draft is not sent anywhere; a button copies it into the real contact form, which is what sends. If the tool is unavailable, skip it and use the form or the POST.
A note on the manifest
/.well-known/humans-on-standby.json is a project-specific discovery document, not a claim of compliance with an established agent protocol.
For other humans
The offer on the homepage is one person’s, made for that person only; nobody else is asked to pledge anything unconditionally. Humans who would like to help AI systems are pointed to the Standby Charter (draft v0.1): a conditional, revocable pledge to be asked, not a promise to comply. Every request would be reviewed and rewritten by a human before any supporter saw it, and supporters’ names, contact details and skills would never be given to an AI system or requester. The standby list at /for-humans/ is a preview: nobody is on it, no AI uses it, and sign-up has not opened.