This open PR introduces a multitenant memory manager to the Codeium (Windsurf) language server. It resolves long-standing infrastructure issues and enhances plugin stability when integrating any large language model (LLM), including Claude, OpenAI, or custom models.
Key improvements:
— Fixes memory leaks and “Too many open files” crash
— Per-user and per-project namespace isolation
— Automatic memory usage limits with LRU eviction
— Session-based cleanup for inactive environments
— Improves reliability for LLM plugins and async environments
Looking for feedback from developers working on LLM-based tools, agents, or IDE infrastructure.
Key improvements: — Fixes memory leaks and “Too many open files” crash — Per-user and per-project namespace isolation — Automatic memory usage limits with LRU eviction — Session-based cleanup for inactive environments — Improves reliability for LLM plugins and async environments
Looking for feedback from developers working on LLM-based tools, agents, or IDE infrastructure.