Rate limits, circuit breakers and model failover for agents' model calls.

children = [
  {WardrikGateway,
   name: MyApp.Gateway,
   fallbacks: %{"anthropic:claude-sonnet-4-5" => ["openai:gpt-5", "zai:glm-5.1"]},
   limits: %{"anthropic" => [requests_per_minute: 4000, tokens_per_minute: 400_000]}},
  {MyApp.AgentRuntime, middleware: [{WardrikGateway, MyApp.Gateway}]}
]

With WardrikCore alone, pass the same entry in an agent's :middleware.

• A model with an open breaker is skipped. The call goes to the next model in fallbacks.
• A call to a provider at its limit is delayed until the provider has capacity.
• A 429 delays calls to that provider for its Retry-After, or one second.
• A 5xx, a 529 or a timeout counts towards the model's breaker. The call goes to the next model.
• After a fallback, the rest of the turn uses the fallback.
• A call is delayed at most :wait. Then WardrikCore.Retry applies.

Options

• :name (atom/0) - Required. The gateway's name.

• :fallbacks - Models to call, in order, when a call to the key's model fails. The default value is %{}.

• :limits - requests_per_minute and tokens_per_minute, by provider, as "anthropic". Providers not listed are unlimited. The default value is %{}.

• :breaker (keyword/0) - When a model is skipped. The default value is [].

  • :failures (pos_integer/0) - Failures that open it. The default value is 5.

  • :within (non_neg_integer/0) - Milliseconds the failures fall within. The default value is 30000.

  • :open_for (non_neg_integer/0) - Milliseconds before one call is let through to test the model. The default value is 30000.

• :wait (non_neg_integer/0) - The longest a call waits, in milliseconds, for room or for a model to come back. The default value is 600000.

Telemetry

• [:wardrik_gateway, :failover] - from, to, reason and agent_id.
• [:wardrik_gateway, :limited] - measures wait, in milliseconds, with model and agent_id.
• [:wardrik_gateway, :breaker_opened] and [:wardrik_gateway, :breaker_closed] - model.