Skip to content
Motir
Start free

InfrastructureMotir Agent Fleet

Your whole plan, worked in parallel.

Motir's hosted agents take every work item that is ready and work them at the same time, each on a fresh machine with its own branch and pull request. Start them yourself, or let Motir Project Manager keep them busy, and the project moves on every front at once.

A few sentences is enough. Motir asks about the rest.

Scale

One machine for every ready work item

Everything that can start, starts. Each run claims its work item, so no two agents ever take the same one, and works it on a machine of its own. A story runs its pieces in the order they depend on each other. The more work is ready, the more runs at once.

Hands-free

Motir Project Manager keeps the fleet busy

Hand the project to Motir Project Manager and it starts a run the moment a work item is ready, picks the model, and brings you only what needs your decision: the pull requests, the designs, the plans.

The right model for each job

Every work item has a difficulty. It picks the model.

The planner gives every task, subtask and bug a difficulty: how hard the work is to get right, not how big it is. A one-line change to lock ordering can be high while a large rename is trivial. The fleet reads it to choose the model, so a rename doesn't pay for the strongest model and a data migration never runs on the cheapest.

  1. TrivialMechanical work with an unambiguous spec: a rename, a copy change, a config flip. Runs on a faster, cheaper model.
  2. LowRoutine work that follows a pattern the code already has, like a new field through an existing form. Runs on a faster, cheaper model.
  3. MediumReal design choices across several files or services, or a bug whose cause isn't known yet. Runs on the strongest model.
  4. HighWork where a subtle mistake is costly: concurrency, security, a migration on live data. Runs on the strongest model.
  5. Your choiceSet your own model for any level in the project, or pick one for a single run, from Claude, DeepSeek, GLM and Qwen.
  6. Pay for what runsBilled in credits for model calls and machine time. Nothing runs, nothing is billed.
How difficulty levels work, and what a task costs on each model

Runs that look after themselves

Problems handled without you

  1. Checks failingA run repairs its pull request on the same branch, up to five attempts.
  2. A run that diedThe work stays on its branch, and a new run continues it without starting over.
  3. A stalled agentA run that goes quiet for 15 minutes is stopped, so a stuck agent never burns credits.
  4. A bug along the wayAn agent that finds a bug outside its own work files it and keeps going.

On the record

Every run, step by step

Each run records its model, every step, what it cost and how each work item ended: implemented, failed, re-planned or skipped.

Works with

Move every front at once.

Bring the idea or the project you have, and let the fleet work everything that's ready.

Start a project