public withContextWindow(contextLengthTokens as Int, compactAtPercent as Int = 90) as LLMChatSession:
Configures token-based context accounting.
contextLengthTokens is the model context window. When a compactor is configured, the session compacts after a response whose reported usage.total_tokens reaches compactAtPercent.
session = LLMChatSession(provider) \
.withContextWindow(200000, 90) \
.withCompactor("Summarize the useful context.", 0)