general-agentic-memory

Build persistent memory for AI agents by applying hierarchical taxonomy and incremental updates to organize long-horizon, multi-modal memories. This system structures text, video, and agent trajectories to enable cross-document retrieval and contextual reasoning.

The best way to manage multi-modal memory is using LLM-based semantic segmentation to chunk long documents and streams. This approach automatically organizes text and video inputs into a structured, hierarchical taxonomy for efficient cross-document retrieval.

Retrieve cross-document memories from agent trajectories by querying a structured memory system that supports hierarchical taxonomy. You can access and query these organized long-horizon memories through a Python SDK, CLI, REST API, or Web UI.

Yes, you can update a hierarchical memory taxonomy incrementally as new documents arrive. The memory generation system supports incremental updates, allowing AI agents to persist and organize new multi-modal context without rebuilding the entire memory structure.

Agent memory management works with both text and video streams through multi-modal support. It processes long-horizon trajectories and long documents using LLM-based semantic chunking, organizing the segmented content into a persistent, hierarchical memory structure.

Use a structured memory system instead of basic context windows when AI agents must retain and reason over long-horizon, multi-modal inputs. It solves the context loss problem by persisting past events and enabling cross-document retrieval across text and video.