Token-Aware Conversation Summarization

What: Intelligent conversation summarization that respects model token limits How to use: Conversation summaries are generated automatically when saving conversations. The system will indicate if summarizing partial conversations:

Summarizing last 5 of 10 messages (~2500 tokens)

Details:

• Dynamically calculates available tokens based on model context window
• Prioritizes recent messages when full conversation exceeds token budget
• Transparently indicates when summaries are based on partial context
• Uses efficient Haiku model with prompt caching for fast summarization
• Generates concise 5-10 word titles for easy conversation browsing