mindspace-dsh-session-memory

Traditional memory systems often focus solely on a single user, making the worldview of long-running agents narrow and prone to disruption when unrelated tasks are injected into conversations or roleplays. The mindspace-dsh-session-memory plugin addresses this by introducing session-isolated multi-person memory. It supports up to five individuals, each with a stable ID, name, information, preference, and revisable relationship background, allowing agents to handle multi-person interactions more naturally without forcibly altering the current task context.

The core capabilities of this plugin lie in its editability and lossless data migration. Agents read and write through get_session_memory and update_session_memory, with memory stored outside the canonical conversation JSONL for isolation. It smoothly migrates data from older versions, converting user profiles and preferences into person one's information, ensuring a seamless upgrade. This design keeps memories external while offering flexible session-scoped controls, avoiding the self-narrating loops that can occur with unsupervised AI-to-AI interactions.

This plugin is ideal for developers building long-term agents that require multi-person interactions, such as roleplay applications, team collaboration tools, or any scenario needing management of multiple real user information. It helps agents maintain task continuity while understanding complex multi-person contexts, leading to more appropriate and coherent responses. If you are creating an AI system that must interact with multiple users simultaneously, this plugin provides structured memory management to enhance the realism and continuity of those interactions.