Kimi Review

Always use Kimi K3.1 Pro at Max reasoning, orchestrated by Cloudroom. The model and effort never change. Only the harness or route may change, and only to get the review running.

Routes

Try these in order. Move to the next one only when the current one fails.

1. --provider pi --model vercel-ai-gateway/moonshotai/kimi-k3.1-pro --reasoning-level max
2. --provider acp-fx --model moonshotai/kimi-k3.1-pro --reasoning-level max (fx uses Vercel AI Gateway)
3. --provider acp-opencode --model vercel/moonshotai/kimi-k3.1-pro --reasoning-level max
4. --provider pi --model openrouter/moonshotai/kimi-k3.1-pro --reasoning-level max
5. --provider acp-opencode --model openrouter/moonshotai/kimi-k3.1-pro --reasoning-level max

Vercel comes first because it keeps no data and does not train on it. OpenRouter serves this model only through Moonshot itself. Vercel routes may incur usage costs.

Pi setup

Pi's built-in catalog lacks this model. Both Pi routes need the entries in references/pi-models.json, placed under providers.<provider>.models in ~/.pi/agent/models.json. If an entry is missing (common in fresh Cloud sandboxes), back up the file, merge the entry in, and confirm with pi --list-models | grep k3.1.

After a models.json edit, Cloudroom's Pi catalog can stay stale for about 5 idle minutes. A stale catalog makes the spawn fail instantly, at no cost. Wait 5 minutes without querying Pi models, then retry once.

Run

Read /room-cli and follow its launch checks. Reuse the current environment, including uncommitted work, and use --parent-self.

A route fails when the spawn errors, room-cli thread show reports a different model or effort, or the first turn errors (gateway errors, auth errors, an empty balance). Stop and archive the failed thread, note its error, and try the next route.

Report a blocker only after every route has failed, with each route's error.