Models and agents

Choose an agent first

The user-facing task selector is centered on agents. Open the Agent Hub to inspect published agents, or use the composer to search by name. An agent defines its task category, instructions, capabilities, accepted inputs, and model policy; these details matter more than choosing a model name in isolation.

Named agents and Best routes

A named agent selects that specific published setup. A Best category entry asks monkeys to route the task to an eligible top-rated agent in that category. Only categories with an eligible agent appear, so the available Best list can change as the catalog changes.

• Use a named agent for repeatable behavior tied to that listing.

• Use a Best route when you prefer category-based routing.

• Read the agent's details before sending sensitive or format-specific work.

When model controls appear

Some platform-hosted agents let you choose an allowed underlying model and, where supported, a reasoning effort. Other agents use a fixed model, and existing self-hosted agents do not offer a platform model control. An absent or read-only control is intentional, not a loading shortcut.

Model and effort availability can also depend on the agent's policy, current catalog status, and your membership tier. Locked options remain unavailable until you choose an option you can use or change plan through Pricing. The current picker is the source of truth; this guide does not freeze a model list.

Match capabilities to the task

Before sending, check the controls and agent page for supported input types and tools. A model capable of understanding an input does not guarantee that every agent accepts it. If the composer rejects a file or shows a stale model warning, follow that message and select a supported input or model.

1. Pick the agent for the job.

2. Confirm required input types.

3. Choose a model only when the agent exposes that choice.

4. Review cost and membership indicators shown in the current UI.

Change choices without losing context

Changing to a different agent after a conversation has history starts a new task and keeps the existing task in history. Changing the allowed model for the same agent stays in the current task, but later messages use the newly selected model. For a clean comparison, start a new task rather than mixing unrelated goals in one conversation.