Skip to content

instructa/switchloom

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

64 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Switchloom

Model routing for coding agents, without hand-writing provider config.

Switchloom turns a small routing policy into repository-local agent roles. You choose which models handle implementation, review, and verification; Switchloom generates the native configuration for your coding agent.

Build a routing setup →

your coding agent (orchestrator)
├── implementer  → fast or cost-efficient model
├── reviewer     → stronger reasoning model
└── verifier     → lightweight verification model

The model selected in your coding agent remains the orchestrator. Switchloom configures the child roles it can delegate to.

Why Switchloom

  • Use the right model for each job. Keep routine implementation and checks on efficient models while reserving stronger reasoning for review.
  • Start from a useful preset. Light, Balanced, and High provide practical defaults; every child role remains editable.
  • Keep setup local to the repository. Switchloom writes project-native configuration without changing global agent settings.
  • Review every change. Preview shows the exact files before apply, and the lifecycle supports status, update, rollback, and uninstall.
  • Use one workflow across hosts. Codex, Claude Code, Cursor, OpenCode, and Pi share the same routing policy while receiving their own native artifacts.

Quick start

The easiest path is the generator at switchloom.ai. Choose a host and preset, copy the generated command, and run it inside the repository you want to configure:

npx switchloom apply --recipe 'sw1_...' --repository .
npx switchloom doctor codex

apply previews the change and asks before writing. doctor checks the local host installation and explains any configuration or reload issue.

To work directly from the CLI:

npm install --global switchloom

switchloom compile balanced --host codex-openai --output routing-bundle.json
switchloom preview routing-bundle.json --repository .
switchloom apply routing-bundle.json --repository .
switchloom doctor codex

Homebrew is also supported:

brew install instructa/tap/switchloom

Presets and roles

Preset Best for
Light Small changes and low-cost routine work
Balanced Everyday implementation with independent review and verification
High Important changes where stronger review matters more than cost

Presets are starting points, not locked bundles. You can change a role's model or reasoning effort, or remove roles you do not need.

Repository lifecycle

switchloom status --repository .
switchloom update routing-bundle.json --repository .
switchloom rollback --repository .
switchloom uninstall --repository .

Switchloom tracks only the files it manages. It refuses unsafe paths and preserves unrelated repository configuration.

Supported hosts

  • Codex
  • Claude Code
  • Cursor
  • OpenCode
  • Pi

Standalone use is the default. Planr integration is optional and does not add a runtime dependency on Planr.

What Switchloom does not do

Switchloom compiles and applies routing configuration. It does not replace your coding agent, make model calls itself, control provider billing, or guarantee that a custom model combination will be cheaper or better. Custom setups should be reviewed and tested in the target repository.

Documentation

License

MIT. See LICENSE.

About

Deterministic model routing for coding agents.

Topics

Resources

Contributing

Security policy

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages