Context Window Management

Overflow Strategies

When a request exceeds a provider's context window:

Strategy	Behavior
route_to_larger	Route to a compliant provider with a larger context window
summarize	Summarize older messages to fit within the window
truncate	Remove oldest messages (preserving system prompt)

Configuration

context_management:
  strategy: route_to_larger
  max_summarization_ratio: 0.5
  preserve_system_prompt: true
  preserve_last_n_messages: 3

Fail-Closed Behavior

If route_to_larger is configured but no compliant provider has a sufficient context window, the request is rejected — it will not route to a non-compliant provider.