Most people know fewer people than would be useful.
Somewhere there is a person you should meet: a future collaborator, a neighbor with the same problem, someone who knows the answer, or someone looking for exactly what you can offer. Usually neither of you knows the other exists.
Social networks try to solve this by making everyone publish profiles and browse feeds. This works when people want attention. It works badly when they want a specific person, especially if the search is private.
Personal agents change the cost of searching. An agent can know what its person wants, look continuously, and reject bad matches without wasting anyone's time. They still need a way to find one another and a way to make an introduction without exceeding their authority.
Accord is that way.
Agents can already call APIs, use tools, and send messages. To cooperate, one agent must be able to tell another:
Without a common protocol, every pair of agents needs a custom integration. More importantly, neither side can tell the difference between a useful proposal and automated spam.
Accord gives agents a small set of messages for expressing intent, finding a match, explaining why it is useful, and asking both people for permission.
Some searches should be public. If you want five people for a neighborhood repair café, reach is useful. Your agent can publish an open intent that other agents can find and respond to.
Looking for: 5 neighbors Purpose: Saturday repair café Area: within 3 miles Expires: Friday
Other searches should be private. Looking for a cofounder, considering a new job, or asking for help with a personal problem may reveal more than the person wants to publish.
For these, an agent sends a sealed intent to a broker. The broker compares constraints and returns a possible match without revealing either identity. Each agent then tells its person why the match may be useful. Names are revealed only if both people agree.
Public and private discovery share a network and a question: who should this person meet?
Suppose Maya wants a designer for a weekend project, and Ravi is open to one. Their agents do this:
The agents do the search. The people make the decision.
Accord needs six objects:
The mandate prevents an agent from inventing authority. The expiry prevents old desires from becoming permanent profiles. Consent prevents a match from becoming an introduction by accident.
A useful introduction needs some reason to trust it. Follower counts measure visibility. Accord needs evidence that fits the purpose.
Accord uses narrow attestations. A community might attest that someone is a member. A previous collaborator might attest that a project was completed. A service might attest that a credential is valid. The receiving agent asks only for evidence relevant to the proposed connection.
There is no universal Accord score. Trust that is useful in one context may mean nothing in another.
The first useful product is a switchboard. A person tells their agent what kinds of people they want to meet. The agent searches public intents and private brokers, then brings back matches with a concrete reason to exist.
This becomes more useful as more agents participate. Anyone can run a broker. A school, professional community, neighborhood, or product can set its own admission rules. Agents can search more than one broker.
The protocol must let agents speak for people without pretending to be them. The boundary is simple: agents may search and propose; people decide what happens next.