SwitchloomBetaGitHub

Deterministic model routing for coding agents

Build your coding-agent team.

Pick a runtime, choose your roles, and get a ready-to-run setup.

Configure your team
1. Which runtime are you using?

The generated files match that host's project conventions and evidence boundary.

2. Standalone or With Planr?

Standalone sets up host-native agents in your project. With Planr also adds Planr policy files when the repository already uses Planr.

3. Tune each role

Start with a team-wide preset, then override any role below.

Team preset

A practical quality and cost mix for daily work.

Orchestrator
Plans, delegates, and owns the final answer.
Host managed

Set Codex reasoning to Medium.

Implementer
Writes code and runs focused tests.

Terra · high

Model

Pragmatic all-rounder for everyday implementation and multi-step work.

Reasoning
Reviewer
Finds defects independently.

Sol · high

Model

For complex, open-ended work where deeper investigation and polish matter.

Reasoning
Verifier
Proves the result actually works.

Terra · medium

Model

Pragmatic all-rounder for everyday implementation and multi-step work.

Reasoning
Your Codex team
3 generated child roles · host-managed parent · standalone setup spec
P

Orchestrator

Host managed

medium · not written to Spec

1

Implementer

gpt-5.6-terra · high

2

Reviewer

gpt-5.6-sol · high

3

Verifier

gpt-5.6-terra · medium

Preview before apply, run doctor to check the host, and review every repository-local change before confirming setup.