PrismMessageMemory class

Conversation memory that trims old turns and can summarise them.

Wraps a List<AiChatMessage> and keeps the most recent turns within a token budget. When the budget is exceeded, the oldest turns are removed. Optionally, a summarise callback is invoked with the removed turns so the caller can store a summary and inject it as a system message.

final memory = PrismMessageMemory(
  maxMessages: 20,
  onSummarise: (removed) => print('${removed.length} turns dropped'),
);
memory.add(AiChatMessage.user('Hello'));
memory.add(AiChatMessage.assistant('Hi!'));

// Pass to the client:
client.streamChat(memory.messages, model: 'gpt-4o-mini');