KokoroMemo
A local long-term memory core for AIRP, AI games, AI table pets and OpenAI compatible clients.
KokoroMemo is a local role continuity engine for AI role-playing (AIRP), AI games, AI table pets, SillyTavern front-end and other OpenAI-compatible clients. It is not another general personal memory assistant, nor is it an enterprise knowledge base RAG. KokoroMemo focuses on three things that AIRP users are really prone to experience: Don't forget: roles can remember names, preferences, boundaries, relationship progress, important agreements and common experiences across conversations. No string: Try to avoid mutual pollution between different roles, different conversations, different world views and different memory banks. Don't forget: temporary plots, jokes, misunderstandings and unilateral fabrications of models will not easily become long-term memories. As an OpenAI-compatible agent running on this machine, KokoroMemo is located between your AI client and the real big model API. The client still requests the chat model as usual, and KokoroMemo is responsible for saving the conversation, maintaining the conversation state, retrieving the long-term memory, and injecting the necessary context into the model. Its goal is not to simply stuff chat records into the vector database, but to use reviewable memory cards, tabular conversation state boards, on-demand recall, local semantic index and retrieval explanation to keep the continuity of character relationship, plot state and playing rules in long-term interaction. Description: 1. Support lazy cat OIDC login and local login. The default account password is admin/12345678. 2. Set the called api key in the service settings, which is optional. The called modelId is consistent with the configured upstream modelId.




