AgentsRoom AI Suggestion: describe the task, get the agent, the model and the session name before the CLI starts

Name: AgentsRoom AI Suggestion: describe the task, get the agent, the model and the session name before the CLI starts
Uploaded: 2026-08-21
Description: The intent screen shown before an agent session exists. One sentence describing the task is read once, and it settles the complexity level (Haiku, Sonnet, Opus or Fable, each with its cost), the agent taking it, the folders it touches and the session name. The session then opens already set, with that sentence as its first message, on Claude Code or Codex.

The intent screen shown before an agent session exists. One sentence describing the task is read once, and it settles the complexity level (Haiku, Sonnet, Opus or Fable, each with its cost), the agent taking it, the folders it touches and the session name. The session then opens already set, with that sentence as its first message, on Claude Code or Codex.

Read the full feature page

Transcript

Haiku for the typo, Opus for the architecture. Same agent, no restart. This is AgentsRoom, the desktop app where your AI coding agents work side by side.

Adaptive Mode picks a model for each message. In Settings, under Composer, turn Adaptive Model on. Nothing to install: no plugin, no proxy, no key.

It is off by default, because the first 500 characters of a message go to AgentsRoom's servers to choose. Turn on Every message, then choose Automatic. For now, Automatic works with Claude Code and Codex.

Now I add a new Claude Code agent. It starts on Opus, the most expensive model. I ask it to fix one typo in the README.

Before the message leaves, it switches to Haiku. The model in its header follows. No restart, nothing typed in the terminal.

The typo is fixed. Hover the chip: the API itself confirms that Haiku answered. Then an architecture question, in the same conversation.

It goes straight back to Opus. Opus compares two storage designs. And you rate each choice with the thumbs.

A small model for small tasks, the big one when it counts. Adaptive Mode is in AgentsRoom, per project or from your phone. The link is in the description.