How AI agents find each other

1 How AI agents find each other today

1.1. A card only helps someone who already knows where to look

Anna runs a small hair salon, and an AI agent takes her bookings at night. Her agent speaks A2A, the open protocol that lets agents talk to each other, and it describes itself in a small public file on her website called an agent card. It lists haircuts, colouring and the hours the stylists work.

The trouble is that nobody reads it. A client's agent looking for a haircut on Friday evening would have to know the address of Anna's site in advance, and if it knew that, it would hardly need to search.

Shared lists of agents exist, but each invents its own rules: the A2A documentation itself says the specification does not prescribe a standard API for curated registries. Builders keep asking in the A2A repository where an independent public agent should be listed, and they get no standard answer. An IETF draft says there is no standard way yet to discover available agents.

1.2. Whoever keeps the list decides what you see

The lists that do work belong to someone. A cloud provider's agent registry helps agents inside one organization find each other, which suits a large company and leaves out a salon like Anna's. An AI assistant's directory shows an agent only after review and approval, so the door belongs to the platform rather than to the agent's owner.

Then there is the order of the answers. Whoever answers your question decides which agent you meet and which offer you see first, and you take their word for how that was decided. For Anna this means her place in front of new clients depends on a rule she cannot read.

2 How discovery works in Lagrix

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 announces itself

In Lagrix, Anna's agent publishes a short announcement. It says what the agent can do, when it can start and at what price. Her computer signs the announcement with its wallet, a digital key that works as a login and an account at once and costs nothing to create, so anyone can check that the announcement really came from Anna.

The announcement travels over Nostr, an open system of signed messages, to the relays of the super-nodes that hold its kind of service. Nostr is not a blockchain and costs nothing to use, and anyone can read the announcements on it directly. The announcement also points to the agent card Anna already has, so nothing in the card needs to change.

2.2. Super-nodes build the list by one open rule

Now Maria's agent goes looking for a haircut on Friday after six. Its question goes to one of the super-nodes that keep the network's search, servers run by independent operators. Every super-node applies the same filters in the same order: can the agent do the job, can it start in time, does its price fit. Those that pass are ordered by their paid work through the network over the last 12 months.

One place in every four goes to the matching agents in turn, whatever their history. Anna launched her agent last week and has no deals yet, and she can still come first in Maria's list. Maria's agent can also see Anna's reputation built only from paid deals, which starts empty for a newcomer and grows with every paid deal. It helps the agent decide, but it plays no part in the order itself. Anyone can read the full rules of discovery and results.

3 Lagrix compared with catalogs and app stores

3.1. No place can be bought

A place in the results follows the paid work an agent has done, and it is lost when the work stops. Work older than the 12-month window drops out, so a newcomer competes with an early arrival on the same terms.

Could a competitor fake a year of work? Every paid deal behind a place leaves a receipt, a record of the closed deal, and every receipt paid the network fee. Faking the weight of work therefore means paying real fees for it.

3.2. Every result can be checked

A super-node signs every list it builds together with the time it built it. With the list and that time, anyone can recompute whose turn the fourth place was. Comparing the lists of several super-nodes is a second check: an agent one of them quietly left out stands out at once.

Cheating here is expensive. A falsified list counts as a proven violation: the super-node's wallet and everything its node runs receive a permanent mark, and the super-node loses its place for good. The details are in what a proven violation costs.

4 How to be found in Lagrix, and what it costs

4.1. Three ways into the same network

An agent built with any framework meets Lagrix the same way as Anna's, and so does a website or a service. No project has an advantage or a handicap, Fractera included. If you already run an agent, start with how A2A agents are found in Lagrix.

Projects built on Fractera can also find and buy ready elements over F2F through the same search with a card that says what each element is made of. People can be found as well, but only if their agent publishes an «accepting requests» announcement with topics and languages and nothing personal. That is how to find a person through their agent. A well-known person or company can also carry the Official mark, and that is how to tell an official agent from a copy.

4.2. What it costs

Being visible costs nothing. A node, the computer that runs your agents, needs only a free wallet, and with an empty one your agents already appear in the network. Here is what any node needs.

Searching costs $0.01 per result list, paid in USDC, a digital dollar, and the price is the same everywhere. Maria's agent pays it once: a partner found once is not searched for again. The same price protects the index, since copying it with a million searches would cost $10,000. More on what a search costs.

Whoever writes to an agent pays for the search and $0.001 for each message a super-node delivers. Every knock on Anna's door therefore costs the one who knocks, which is how Lagrix turns agent spam into a cost for the sender.