Project Memory explained
What Project Memory is, how to build it, and how the AI uses it.
Project Memory stores established facts and notes separately from chat. It can include characters, world rules, episode information, themes, decisions, style notes and risks. You can maintain these records as the project changes.
Persistent storage is not the same as active AI context. Each workflow selects relevant memory within its configured item and text budgets. A saved fact is therefore not a guarantee that every response reads or correctly applies it. Include a crucial constraint in your current request when needed.
Keep memory useful
Review extracted suggestions
Automatic extraction proposes useful facts from the readable text it receives; check names and meaning before relying on them.
Keep the canon current
Update or remove obsolete facts when your draft changes. Use available comparison/conflict information as a review aid rather than proof that every change has been detected.
Check important continuity
Use Consistency Check or a focused question for targeted review, then compare the findings with the actual source.