Stateless Single Session Summarization Without Persistence Or History

via “infinite memory engine with recursive conversation summarization”

🪐 The ultimate starter kit for AI IDEs, Claude code，codex, and other agentic coding environments.

Unique: Uses recursive hierarchical summarization (conversation tree structure) rather than sliding windows or vector-based retrieval to manage long conversation histories. Summaries are generated by LLMs rather than extractive methods, preserving semantic meaning while reducing token count. The system maintains a tree structure where parent nodes are summaries of child nodes, enabling multi-level compression.

Unlike sliding window approaches (which lose old context entirely) or vector-based memory retrieval (which requires semantic search), Antigravity's recursive summarization preserves the full conversation structure while compressing token usage. This approach is more transparent and debuggable than vector-based methods, though potentially less efficient for very long conversations.