PerchAI
How it works

Perch Roost

Roost is Perch's routing engine. It picks the right model for each task and races across providers for speed and reliability.

Roost is the engine that decides which model handles your request. You do not choose a model on every message. Roost picks one for you, and you can still pin a specific model by hand when you want to.

The short version

No single model is best at everything. Roost matches each task to a capable model, so you get good results without shopping for the right one yourself.

Smart routing

Your request
A question or task on any surface.
Roost
Matches the task to a capable model, then races providers so a slow or failed one never blocks you.
Provider A
Raced
Provider B
✓ First good answer
Provider C
Raced
Your answer
The chosen model handles the task and stays consistent from start to finish.

When you send a request, Roost selects a model suited to the task and runs it. Two things happen behind that choice:

  • One driving model per task. The model chosen at the start of a task stays fixed for that task, so the work stays consistent from beginning to end rather than drifting as it switches models mid-job.
  • Raced for speed and reliability. Roost can run more than one provider at once and take the first good answer. If one provider is slow or unavailable, another responds, so a single bad provider does not leave you stuck.

The result is that you spend your attention on the work, not on picking and babysitting models.

Images depend on the model

A model that sees images gets your screenshots, photos, and scanned pages directly. A text-only model does not: no second model reads an image for it, so it works from text and tells you when an answer depends on an image it could not read. On Roost, whether a tier sees images depends on the model behind that tier.

See Images and scans for which models see images and what happens when one cannot.

Picking a model yourself

Roost runs automatically by default, which is the right choice for most work. When you want direct control:

  • Starter runs on a rotating list of frontier models, subject to availability. Roost picks from it, and you can also pin any current Starter model by name on Desktop and the CLI.
  • Pro adds Gemini 4 Hydrogen, Gemini 4 Carbon, Gemini 4 Argon, Claude Opus 5.5, GPT-6.1 Sol, GPT-6 Luna, Claude Sonnet 5.5 and Claude Haiku 5.5, and lets you pin any Pro model by hand, while still falling back to Roost whenever you leave it on automatic.
  • Ultra includes everything in Pro and adds Claude Fable 5.2 and GPT-6 Astra, with Claude Fable 5.1 still available beside them. You pin these by hand; Roost never picks them.

In Perch Desktop, the choice belongs to the chat: each chat keeps its own Roost tier or pinned model, and its own effort level. See Model and effort per chat.

See the full model list for what each plan includes, the Pro plan guide for the separate Roost and manual usage limits, and Pricing to compare plans.

Where Roost fits

Roost sits underneath everything Perch does. The read-first loop decides what work to do; Roost decides which model does it.