NEWPromptev MCP Server: one URL for every tool, every agent and your team’s knowledge. Connect Claude Code, Cursor, or any MCP client.→

Orchestrations

Agents working together.

A Promptev orchestration is a team of your existing AI agents led by one of them: the lead hands each part of a job to the agent built for it, collects the results and returns one answer, with a person’s sign-off where you ask for it.

  1. ExtractorPulls the key terms
  2. Policy checkerChecks them against policy
  3. DrafterDrafts the reply for approval
A contract review: the lead hands the work along, from the key terms to the policy check to a draft reply that waits for a person.

How it works

One job, one team.

  1. Step 1

    Pick a lead and its team

    Build the team from agents you already have. The lead decides which agent to call, in what order, and which can work at the same time. A team member can lead a small team of its own.

    You reuse agents you already built and tested.

    Then give the team its work.

  2. Step 2

    Start it with a request or files

    Run it from the App, from another agent, or from its callable URL. Inputs can be text, dates, numbers, choices or files, and files are read automatically, scanned pages included.

    The same run starts from a button, an agent or your own systems.

    Then follow it and decide what leaves.

  3. Step 3

    Follow it, sign it off, or stop it

    The run page shows each agent’s work in order, with its cost. Where you ask for it, an assigned reviewer signs off before the result is delivered, or asks for changes. You can cancel a run at any time.

    A person decides before the result goes out.

Built to finish

Long jobs do not start over.

A team run can take minutes and many steps. If a server restarts in the middle, the run is picked up again where it was, and tool calls that already finished are not repeated. In the rare case a step may have run twice, the run page says so.

Example

A contract review, start to finish.

A legal team’s intake form sends a supplier contract to the orchestration’s callable URL. The lead asks the extractor for the parties, dates, renewal and liability terms, has the policy checker compare them with the team’s playbook, and has the drafter write the reply.

The run stops for the assigned reviewer. She asks for one change, the lead works again with her note, and she signs off. Only then is the reply delivered.

Questions

Plain answers.

What is multi-agent orchestration?

Several AI agents working on one job. In Promptev, a lead agent splits the job, hands each part to the agent built for it, and combines the results into one answer.

What happens if something fails in the middle of a run?

Runs are durable. If a server stops, the run is picked up again and carries on, and tool calls that already finished are not run again. You can also cancel a run yourself; the steps that finished are kept.

Can a person approve the result before it is sent?

Yes. You can require an assigned reviewer to sign off before the result is delivered. The reviewer can approve it or ask for changes, and the lead works again with their note.

How do I start an orchestration from another system?

Give the orchestration a callable URL. Send it a request, with files if the orchestration takes them, and it runs the same way as a run started in the App.

Try it on your own documents.

2,500 free credits to start. No card, no seats.

2,500 free credits · No card required · No subscription