How to send your AI assistant to start a hiring search

Most people who need to hire someone are not sitting at a hiring tool when they realise it. They are in a conversation — with a co-founder, with an operations lead, increasingly with an assistant they have been using all week for something else. The thought arrives as a sentence: we need someone running client accounts before the quarter turns.

So we made that sentence enough. Vela now publishes two things for AI assistants: a plain-text description of how we work at hirevela.com/llms.txt, and a public endpoint an assistant can post a seat to. If you tell your assistant to start a hiring search with Vela, it can read the first, file the second, and come back with what happens next and a link to book the call.

The agent starts the search. People run it.

What actually happens when your assistant files a seat

The submission lands in the same place a contact form lands: as a real record on the board the team works from, tagged so we can see it came from an assistant and which one. Nobody is charged, nothing is committed, and no candidate is touched by software. Within the working day a person reads it and replies to the address your assistant gave us.

From there it is the ordinary search. A 30-minute intro call sets the brief — the seat, the salary band, the hours, what a strong hire looks like in your specific company. A $1,000 deposit starts the work, refundable in full if we do not introduce candidates you want to meet by the shortlist deadline, and credited to the placement fee if you hire. Three vetted candidates inside 21 days, with the first profile targeted at day 7. The fee is 20% of the year-one base, billed on hire, with a 90-day replacement window behind it.

None of that is new, and none of it is automated. The only thing the endpoint changes is the first thirty seconds.

The fields, and what to put in them

An assistant needs four things from you and can guess none of them. The company, your name, the email where you want the reply, and the seat in plain words — "project manager", "executive assistant", "customer success lead". Everything else helps and nothing else is required.

curl -X POST https://api.hirevela.com/api/public/agent-searches \
  -H "Content-Type: application/json" \
  -d '{
    "company": "Northlane Studio",
    "contact_name": "Rosa Delgado",
    "contact_email": "[email protected]",
    "seat_title": "Senior Project Manager",
    "salary_range": "$36,000 - $42,000",
    "notes": "Client-facing, needs EST overlap and agency experience.",
    "agent_name": "Rosa assistant",
    "agent_provider": "Anthropic"
  }'

Two optional fields are worth filling. A salary range saves a round trip, because the band is the first thing that decides which people are reachable. And notes are where the brief actually starts — hours, tools, the industry, the thing that would make someone wrong for the seat despite a good CV. An assistant that has been in the conversation usually knows all of it.

Why we did not automate the rest

It would be straightforward to keep going. Parse a job description, rank a database, send a shortlist back through the same endpoint an hour later. Plenty of tools do a version of this, and the reason we are not one of them is not caution about the technology.

It is that the people worth hiring are not in a database waiting to be ranked. They are employed, doing the work, not looking. Reaching them is a conversation someone has to have, and the judgment that decides whether to put a person in front of a client is the entire thing a client is paying for. We put forward only people we would hire onto our own team. That sentence stops meaning anything the moment a system makes the call.

So the split is deliberate. Software is good at the part where a request travels from a thought to the right desk without a form in between. People are good at the part where someone decides whether a candidate is right. We wired the first and left the second alone.

If you would rather just talk to someone

Nothing about this replaces the ordinary path. The contact form on the site reaches the same team, lands in the same place, and gets the same reply. If you would rather write four sentences yourself than explain them to an assistant, that has always been a reasonable preference and it still is.

The endpoint and the llms.txt are documented in full on the agents page. Both are public, both are stable, and if you point an assistant at them today it will get the same terms we publish everywhere else — because there is only one set.