@0xkobold/pi-learn

This Skill addresses the issue of AI agents losing memory across session interactions. In practical development, agents often fail to remember user preferences, historical conversations, or project context, resulting in each conversation starting from scratch and a fragmented user experience. Pi-Learn enables AI agents to possess persistent, retrievable "memory" by providing structured memory storage, reasoning, and context assembly, thereby supporting more natural and intelligent continuous dialogues.

Usage is straightforward: simply install Pi-Learn as a Skill into your AI agent, then use the learn_add_message tool to store conversation messages or learn_add_observation to log raw observations. Subsequently, the AI agent will automatically perform reasoning (learn_reason_now) and dream processing (learn_trigger_dream) to generate user profiles, conclusions, and summaries. When context is needed, calling learn_get_context retrieves a complete memory snapshot for the current user or project.

It is particularly suitable for development teams building agents that require long-term user relationships or ongoing project collaboration—especially those already using the Pi Agent framework or developers looking to upgrade their AI agents from "one-time tools" to "memory-capable partners." Whether for chatbots, personal assistants, project managers, or educational tutoring applications, all can benefit.

For initial setup, it is recommended to initialize basic user information via learn_update_peer_card and periodically run learn_prune to trigger memory cleanup policies and prevent data bloat. Note that it primarily relies on local storage; for handling massive user volumes, external database scaling may be necessary. Additionally, dream reasoning is an asynchronous background task, requiring the agent to have sustained runtime to take effect.