Multiplet and a stack you build

# Spend the run   on your problem.

Multiplet operates the inference, harness and environments behind the fleet, so your project gets the attention.

01 / Inference 

## Inference built  for a working fleet.

Hundreds of agents repeatedly use project context, instructions and tools. Multiplet’s inference and harness are optimized together for high cache reuse across long runs.

A complete run includes the model service. When assembling your own, model calls, cache behavior, request capacity and scheduling become part of the system you operate.
[Explore inference and environment costs ](/compare/costs/) 

02 / Harness 

## Keep the investigation  moving through long tasks.

Multiplet uses Affordance to manage context, tool execution and working state. Context folding and RLM-style environments let agents work with material beyond a single active context window.

The harness is integrated with the environments and model serving. An open-source release of Affordance is planned.

03 / Operation 

## We run the fleet.  You choose the work.

Environment provisioning, shared files, agent messages, usage records, recovery and returned artifacts belong to one Multiplet run.

Building on public APIs gives you control over the stack. Multiplet gives you that stack as a service, with one place to manage the work and retrieve what it produces.
[See the handoff from your existing agent ](/how-it-works/) 

## 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/)