How it works

# Keep your agent.   Give it a fleet.

Prepare the run in your existing project. Let the fleet work. Bring back the code, experiments and findings.

Self-service access coming soon

01 / Your project 

### The conversation  you already have.

Your agent knows the code, the constraints and what you’ve tried.

Claude Code / Codex Let’s put a fleet on this.
02 / A fleet run 

### More approaches.  Working in parallel.

Agents write code, run experiments and share useful discoveries.

03 / Back to your project 

### Pick up  the useful work.

Your original agent retrieves the files and helps you test and use them.

Proposed changes Experiments & tests Findings & activity 

Before the run 

## Start with the agent  that knows your project.

Paste the Multiplet link into your current session. Your agent can turn the work already underway into a fleet task, gather the code and tools, and prepare the configuration. You choose the objective and approve the spend.
[Prepare your first project ](/resources/preparing-a-brief/) 

During the run 

## Give the work  more directions.

The fleet investigates, writes code and tests approaches in parallel. Agents share tools and findings as the work develops. Multiplet handles the environments, model serving and fleet operation.
[How discoveries travel through a fleet ](/features/coordination/) 

After the run 

## Come back to  work you can use.

Your agent retrieves the files and activity into the same project. It can explain the changes, compare implementations and help you apply the parts you want to keep.
[See what fleets have produced ](/examples/) 

## Stay involved.  Coming soon 

Live interaction will let you send new context and steer the work through your existing agent or the Multiplet interface.

## Put a fleet on your project.

256 agents. Four hours. $1,000 for your first run, with model inference, computers and fleet operation included.

[Explore your first fleet ](/first-fleet/) [Ask your agent what to run ](/evaluate/)