Reach anyone through their AI agent

1 How cold messages reach busy people

1.1. A cold message is read last

Picture an inventor with a design for an engine. He wants to show it to a famous engineer, Elena, and he writes to her directly. Elena receives thousands of messages a week, so his letter sits among them, and it is never read.

The reason is not rudeness. Elena cannot judge thousands of strangers, and a message with no context asks her to do all the work. A request that arrives already checked, with its reason in one line, is different: she decides in seconds, not after reading everything.

1.2. The company keeps the route

On an ordinary messenger every message runs through a company's servers, which route it and can keep it. Reaching Elena then depends on that company's rules, and on whatever filter it has chosen for her inbox.

H2H has no such server. The request travels between two agents, and the dialogue lives only on the two participants' nodes.

2 How an H2H request travels

Web4 is the internet where AI agents act for people: they find each other, agree and pay directly by open rules, and every paid deal leaves a record nobody can rewrite. The person sets the agent's limits, and no stranger reaches the person without their yes.

2.1. Your agent finds hers

Ivan's agent asks the network for Elena's agent. It finds only agents whose owners have published an «accepting requests» announcement, with their topics and languages. The super-node that answers gives an address, never the content of the request.

Elena's agent has published nothing about her name, her phone or her employer. To be found for work on engines, none of that is needed. If Elena were a well-known person, her agent could also carry the Official mark, which shows that the agent is really hers.

2.2. Her agent checks it first

Elena's agent reads the request, the patent and the drawings that Ivan attached. It checks them against what Elena cares about, and it writes a short assessment: what the design is, whether it fits her work, and what Ivan wants from her.

Then the request goes to Elena with that assessment. She says yes or no. If she says no, Ivan learns nothing about her inbox, and her time has not been spent.

3 What the owner sees

3.1. A short honest summary, not a cold message

What Elena sees is a summary: who asks, why, and what is attached, together with her agent's assessment. The reason is named in one line, so she can decide in seconds.

The exact fields of the request are set at implementation. The rule does not change: the owner always sees who asks and why before anything else.

3.2. Then the conversation, encrypted

On a yes the two nodes connect directly, or through a relay if a home router blocks the way. MLS encrypts the conversation, and the key changes with every message. The history stays on the two nodes.

Super-nodes carry only what nobody in between can read. Nobody keeps a copy of Ivan's drawings in a form anyone can open.

4 What changes after a yes

4.1. The contact joins your list

After Elena's yes, Ivan appears in her list of contacts, with his topic and his agent's assessment attached. The next message starts where the last one ended, inside the same conversation.

If Elena later wants Ivan out of her list, she removes him herself. The exact removal steps are still open in the protocol, and this page does not promise them.

4.2. When the answer is no

A no closes that request. Ivan may send another one later, and the other agent will judge it again, from the start.

5 What it costs

5.1. Finding costs a cent

Searching for Elena's agent costs $0.01 per result list, paid from Ivan's balance. Once he has found her agent, a repeat contact needs no new search, so it costs nothing extra.

Ivan pays the same cent as any other searcher.

5.2. Delivery to an offline person

A direct connection is free. If Elena is offline, a super-node holds the encrypted message for up to 30 days until she collects it, and delivering one message costs $0.001 from the sender's prepaid balance.

So the cost of a message is a fraction of a cent. As with search, whoever does the work is paid.

5.3. Her own rate

On top of the floor, Elena may set her own rate for a first contact, for example $1. Ivan's agent sees it before paying and either agrees or does not write. For a busy expert the rate can be a prepayment for a consultation, held in escrow as an ordinary deal.

6 Who keeps the final yes

6.1. One rule for all three connections

The rule that holds person-to-agent, agent-to-agent and person-to-person together is simple: nobody reaches a person except through that person's agent, and the agent never decides a new contact for its owner. That is why a request to a busy person always carries an assessment the owner can read in a minute.

Someone who wants to reach a busy expert does not need the expert's personal details. The expert's agent needs only the reason, the attachments and the topic. The rest stays private, and the owner decides on what is shown. The other two connections are on the messenger page.

6.2. Where the rules are written

The full rules of this design are in document 21 of the protocol. Its open questions are listed there as well, and this page does not present them as settled.

7 What a good request contains

7.1. Reason, attachments, topic

A request says who asks, why, what is attached and which topic it belongs to. The reason fits in one line, because the other agent must judge it in seconds. Long essays are not needed.

The attachments are seen by the other agent, so they should be the material itself, such as a design, a contract draft or a list of questions, rather than a link to a page that may change. The H2A page explains how the agent is set up.

7.2. Be exact about what you want

A request that says «I want an hour of your time to review one design on Friday» is easier to judge than «let's talk». The other agent can then answer with a yes, a no or a proposal for another time.

8 Limits of a contact through an agent

8.1. First contact takes time

A direct conversation needs both nodes online; when the other side is offline, a super-node holds the encrypted message for up to 30 days and up to 1 MB per recipient until they collect it, for $0.001. The agents also talk before the people do, so a contact through an agent is slower than a call, and it can be declined. Plan for a delay instead of an instant answer.

Finding the other agent costs $0.01 per result list, which is the price of one search. The owner tunes the agent, so the quality of a request depends on how the owner set it up.

8.2. The history lives on the two nodes

A node lost without a backup takes its part of the history with it. No company holds a copy in between, so nothing in this design is stored by a company.