The routing config

Two files in ~/.omp/agent/ do it. config.yml maps roles to models and holds fallbacks and folder rules. models.yml adds providers omp doesn’t know. Model ids look like provider-id/model-id, and any can take a thinking-effort suffix such as :high or :minimal.

Roles

Also vision, designer and task. Set them in config.yml:

modelRoles:
  default: claude-opus-4-6
  smol: "@default"
  slow: openai/gpt-5.5

"@default" reuses another role’s model. Start with frontier on default, slow and plan, a small model on smol and commit, and leave it for a week. Switch with /model mid-session, cycle the active role’s model with Ctrl+P, or pass omp --model … --smol … --slow … --plan ….

models.yml: add an endpoint

The README’s example for an OpenAI-compatible server:

providers:
  spark:
    baseUrl: http://192.168.10.223:8000/v1
    api: openai-completions
    apiKey: dummy
    models:
      - id: minimax-m3
        name: MiniMax M3
        contextWindow: 100000
        maxTokens: 32000

Check it with omp models spark, then route to it (smol: spark/minimax-m3). Ollama, LM Studio and llama.cpp don’t need this, since omp finds them on their default ports. api also accepts anthropic-messages, openai-responses, google-generative-ai and others.

Fallback chains

On by default (retry.modelFallback). A model that keeps returning 429 hands its work to the next entry in its chain and comes back after the cooldown. From the settings docs:

retry:
  fallbackChains:
    default:
      - anthropic/claude-opus-4-5
      - openai/gpt-5.5
      - google/gemini-3-pro
    smol:
      - openai/gpt-5.5-mini
      - anthropic/claude-haiku-4-5

Keys can be roles, exact provider/model-ids or provider/* wildcards. google-vertex/* means the same model through another provider. Unknown models and broken chains show up as warnings at startup, so read them.

Rules by folder

enabledModels and disabledProviders accept path-scoped entries, so one global config can set different rules per folder:

enabledModels:
  - claude-sonnet-4-5            # applies everywhere
  - path: ~/work/high-context
    models:
      - anthropic/claude-opus-4-5

disabledProviders:
  - paths:
      - ~/clients/acme
    providers:
      - openai

The usual use: allow only local models in a client’s repo, so their code never leaves your machine. An entry applies when your working directory is at or below the path.

Spend

Run the @oh-my-pi/omp-stats dashboard after a week to see your spend per role, in place of the simulator’s toy units. If one key’s limit holds you back, omp can rotate several keys per provider.