Model Router

alpha 0.3.0-alpha.4

Published release candidate for @wf/model-router@0.3.0-alpha.4. Follow the registry guide and review compatibility before installing.

Join the alpha Sign in with Google, then add your name. We review alpha access requests.

# Public configuration; relative paths are based on this file.
credentials:
  nvidia: nvidia-api-key
  openrouter: openrouter-api-key
settings:
  mode: auto
  maxEffort: auto
# Total attempts per model, including temporary failures and invalid answers.
recovery:
  maxAttempts: 3
  baseDelayMs: 250
  # One correction per model, then an eligible fallback.
  maxQualityAttempts: 2
  # Shared by classification, corrections and every fallback.
  timeoutMs: 120000

# Optional: official CLIs signed in to subscriptions on the trusted host.
# The app must authorize these references for its current user.
# subscriptions:
#   codex:
#     authorization: my-codex-subscription
#   claude:
#     authorization: my-claude-subscription

# Switchyard is on by default, using an eligible model at low effort.
# One deadline covers the entire tier question, including fallback lanes.
# classifier:
#   enabled: true
#   timeoutMs: 20000
#   operations: [generation, validation]  # Optional: other tasks use deterministic sizing.
#   decisionUrl: http://localhost:8080/decision  # Optional existing Switchyard API.