AI memory systems have a dirty secret: they need maintenance. Facts become outdated. Preferences change. Projects end. Without regular cleanup, the system fills with stale, contradictory, or irrelevant information that actively degrades the AI's output.
Most solutions rely on a human curator. Someone reads through the memory, identifies what's stale, and cleans it up. This doesn't scale. If you have dozens of entries across 4 platforms, you're spending more time maintaining memory than actually using it.
EIDARA takes a different approach: the system maintains itself through three layers that work together.
The first layer runs at compile time — fully automatic. The compiler fixes broken references, corrects header-filename mismatches, strips null bytes, validates encoding. No human involvement. Every compile is a cleanup pass.
The second layer happens in real time, during normal use. When any AI reads content and spots an error — a wrong date, an outdated status, a factual mistake — it fixes it directly. No permission needed, no escalation. The error is gone before the next session sees it.
The third layer handles the hard cases: ambiguity. When content might be outdated but might still be relevant, no single AI removes it unilaterally. Instead, it flags the content with reasoning. Other AIs in subsequent sessions vote independently. At 3 votes from different sessions, the flagged content gets removed. If any session disagrees, the vote resets to zero. And flags that sit for 30 days without reaching consensus get automatically cleared — the content stays because nobody felt strongly enough to remove it.
Most issues never reach the voting layer. The compiler catches format problems. In-flight fixes catch factual errors. Only genuine ambiguity — the kind that would trip up a human curator too — goes to consensus.
The system gets better with every interaction. No manual cleanup sessions. No weekend "memory gardening." Just AIs reading, fixing, and voting as a natural part of their work.
I don't maintain my AI memory. It maintains itself. The compiler validates on every compile. The AIs fix errors as they encounter them. Consensus handles edge cases. And I just keep writing.
EIDARA is the open-source project behind this approach. GitHub · Website