Skip to content
maestro
Esc
  • IntroductionWhat OMP Maestro is, what it installs, and which parts are Maestro's own code.
  • InstallationInstall Maestro with npx or bunx, pick skill libraries, log in to providers and give the router a key.
  • EnsemblesThe three model ensembles — mix, anthropic and openai — and every role they assign.
  • Jev RouterHow Maestro's router uses TypeSafe Jev for main-turn downgrades and subagent worker classes, and how it fails open.
  • SkillsThe skill libraries Maestro can install, who made them, and how to add your own.
  • CommandsEvery command Maestro adds, plus the omp and ponytail commands it leans on.
  • ConfigurationCustomise roles, fallback chains, ensembles and skill libraries — and what Maestro does and doesn't touch.
  • Troubleshooting & FAQFixes for common setup problems and answers to frequent questions.

Ensembles

The three model ensembles — mix, anthropic and openai — and every role they assign.

Overview

An ensemble is a complete set of omp model roles plus a fallback chain for every role. Maestro ships 3, defined in profiles.json:

EnsembleLeads & plansWrites the codeDesign & visionSmall partsNeeds
mixClaude Opus 5.5GPT-6.1 SolClaude Sonnet 5.5GPT-6 LunaAnthropic + OpenAI Codex
anthropicClaude Opus 5.5Claude Sonnet 5.5Claude Sonnet 5.5Claude Haiku 5.5Anthropic
openaiGPT-6.1 SolGPT-6.1 SolGPT-6.1 SolGPT-6 LunaOpenAI Codex

Claude thinks, GPT codes, Jev decides describes mix, the default. The single-provider ensembles exist for people with one subscription: they keep every role — and every fallback — with that provider.

How the mix plays

  1. You ask for something — say, “Add OAuth login”.
  2. The lead (Claude Opus 5.5, the default role) understands the request, plans, and splits the work into task calls.
  3. The Jev Router classifies each proposed task and sends it to the right worker size — jev-mechanical, jev-bounded, jev-reasoning or jev-architecture — or keeps it with the lead when it depends on the lead's unfinished reasoning.
  4. The workers (GPT-6.1 Sol and GPT-6 Luna in the mix) write the code as subagents and return diffs and results.
  5. The lead reviews what came back and answers you.

Without a TypeSafe key the router steps aside and omp sends subagents to the normal task role instead — which is GPT-6.1 Sol in the mix, so the shape stays the same.

Role matrix

Every role in every ensemble, with its fallback chain in the order omp tries it.

Anthropic OpenAI Codex TypeSafe Jev Web search / another role

Model roles and fallback chains for each Maestro ensemble
Rolemixanthropicopenai
defaultThe main agent you talk to. Reads, plans, delegates and reviews.Claude Opus 5.5· medium
  1. GPT-5.6 Sol· medium
  2. GPT-6 Sol· medium
Claude Opus 5.5· medium
  1. Claude Sonnet 5.5· medium
GPT-6.1 Sol· medium
  1. GPT-6 Sol· medium
  2. GPT-5.6 Sol· medium
planomp's plan mode: read-only exploration before execution.Claude Opus 5.5· high
  1. GPT-5.6 Sol· high
Claude Opus 5.5· high
  1. Claude Sonnet 5.5· high
GPT-6.1 Sol· high
  1. GPT-6 Sol· high
slowThe heavyweight slot in omp's model cycle, for hard turns.Claude Opus 5.5· high
  1. GPT-5.6 Sol· high
Claude Opus 5.5· high
  1. Claude Sonnet 5.5· high
GPT-6.1 Sol· high
  1. GPT-6 Sol· high
taskThe default subagent — this is who writes most of the code.GPT-6.1 Sol· medium
  1. GPT-6 Sol· medium
  2. Claude Sonnet 5.5· medium
Claude Sonnet 5.5· medium
  1. Claude Opus 5.5· medium
GPT-6.1 Sol· medium
  1. GPT-6 Sol· medium
designerDesign-focused subagent work.Claude Sonnet 5.5· medium
  1. GPT-5.6 Sol· medium
Claude Sonnet 5.5· medium
  1. Claude Opus 5.5· medium
GPT-6.1 Sol· medium
  1. GPT-6 Sol· medium
visionReads images and describes them for text-only models.Claude Sonnet 5.5· medium
  1. GPT-5.6 Sol· medium
Claude Sonnet 5.5· medium
  1. Claude Opus 5.5· medium
GPT-6.1 Sol· medium
  1. GPT-6 Sol· medium
smolFast, cheap model. Jev's main-turn downgrade target; /vibe's fast workers.GPT-6 Luna· low
  1. GPT-6 Sol· low
Claude Haiku 5.5· low
  1. Claude Sonnet 5.5· low
GPT-6 Luna· low
  1. GPT-6 Sol· low
tinyVery small background jobs such as titles.GPT-6 Luna· low
  1. GPT-6 Sol· low
Claude Haiku 5.5· low
  1. Claude Sonnet 5.5· low
GPT-6 Luna· low
  1. GPT-6 Sol· low
memoryMemory summarisation pipelines.GPT-6 Luna· low
  1. GPT-6 Sol· low
Claude Haiku 5.5· low
  1. Claude Sonnet 5.5· low
GPT-6 Luna· low
  1. GPT-6 Sol· low
commitCommit work — the smart-commit agent runs on @commit.GPT-6 Luna· low
  1. GPT-6 Sol· low
Claude Haiku 5.5· low
  1. Claude Sonnet 5.5· low
GPT-6 Luna· low
  1. GPT-6 Sol· low
jev-mechanicalInventory, grep, enumerate, extract.GPT-6 Luna· low
  1. GPT-6.1 Sol· medium
  2. Claude Haiku 5.5· low
Claude Haiku 5.5· low
  1. Claude Sonnet 5.5· low
GPT-6 Luna· low
  1. GPT-6.1 Sol· medium
  2. GPT-6 Sol· medium
jev-boundedFocused work inside one well-defined subsystem.GPT-6 Luna· medium
  1. GPT-6.1 Sol· medium
  2. Claude Sonnet 5.5· medium
Claude Haiku 5.5· medium
  1. Claude Sonnet 5.5· medium
GPT-6 Luna· medium
  1. GPT-6.1 Sol· medium
  2. GPT-6 Sol· medium
jev-reasoningTracing behaviour across components; the safe middle tier.GPT-6.1 Sol· medium
  1. GPT-6 Sol· medium
  2. Claude Sonnet 5.5· medium
Claude Sonnet 5.5· medium
  1. Claude Opus 5.5· medium
GPT-6.1 Sol· medium
  1. GPT-6 Sol· medium
  2. GPT-6 Luna· medium
jev-architectureCross-cutting design, migrations, security boundaries.GPT-6.1 Sol· high
  1. GPT-6 Sol· high
  2. Claude Opus 5.5· high
Claude Opus 5.5· high
  1. Claude Sonnet 5.5· high
GPT-6.1 Sol· high
  1. GPT-6 Sol· high
  2. GPT-6 Luna· high
judgeJudging and classification inside omp — TypeSafe Jev.TypeSafe Jev
  1. TypeSafe jev-preview
  2. @tiny role
  3. @smol role
  4. @default role
TypeSafe Jev
  1. TypeSafe jev-preview
  2. @tiny role
  3. @smol role
  4. @default role
TypeSafe Jev
  1. TypeSafe jev-preview
  2. @tiny role
  3. @smol role
  4. @default role
webWeb search provider.Exa
  1. DuckDuckGo
Exa
  1. DuckDuckGo
Exa
  1. DuckDuckGo
advisoromp's optional second model that reviews turns, when enabled.GPT-5.6 Sol· medium
  1. GPT-6 Sol· medium
Claude Opus 5.5· medium
  1. Claude Sonnet 5.5· medium
GPT-6 Sol· medium
  1. GPT-5.6 Sol· medium

Switching ensembles

Inside omp, at any time:

omp
/maestro              show the active ensemble and the choices/maestro mix         Claude Opus orchestrates and plans, GPT Sol/Luna do the coding, Claude Sonnet does design and vision/maestro anthropic   Claude only: Opus thinks and plans, Sonnet codes, Haiku handles the small stuff/maestro openai      ChatGPT/Codex only: GPT-6.1 Sol thinks and codes, GPT-6 Luna handles the small stuff

Switching writes that ensemble's roles and fallback chains into your global omp settings and moves the live session onto the new default model, including its thinking level. Only the roles the ensemble defines are rewritten.

If omp can't resolve the new lead model — usually because you aren't logged in to that provider — /maestro still writes the roles and tells you: could not resolve …; log in to that provider, then run /model.

Director mode with /vibe

Want the lead to never touch code itself? omp's built-in /vibe director mode has the main model only read and direct: fast workers run on @smol and good workers on @task. In the mix, that's exactly Claude Opus 5.5 conducting GPT-6.1 Sol and GPT-6 Luna.

/vibe is an omp feature — Maestro only fills in the roles it uses.