I experienced this friction. Maintenance of memory across projects, across different LLMs, is a recipe for frustration and disaster.

This MCP tool I made might help elicit desired behaviour from Turn 0: https://github.com/turnzero-ai/turnzero