2026-05-17
Two months ago I shipped EIDARA v1.0 — a constitution and a compiler that gives any AI persistent memory across sessions. I've been running it on my own work every day since. Today I'm shipping v2.0 with the lessons that came out of that, and pointing at where it's heading.
v2.0 is the same EIDARA, just self-correcting. The big themes:
The system now asks you to maintain it, instead of waiting silently. Previously, when the Librarian (the hygiene pass) was overdue, only the maintainer noticed. Now every AI that reads BRAIN.md sees the status in the header and prompts you to run it. Same pattern for missing neuron summaries that silently inflate token costs.
The compiler refuses to be fooled. A new anti-tampering check cross-references the Librarian's "last run" timestamp against the changelog. If the file says "ran today" but no real Librarian activity exists for today, the compiler flags it as suspicious. End of silent decay.
The constitution learned a hard lesson about sandbox filesystems. Working in environments with FUSE mounts (dev containers, sandboxes, Cowork-style tools) can serve stale cached reads that look like file corruption. v2.0 codifies this: never roll back from a single sandbox-side read.
Smaller things: session-closing summaries are capped at 500 words, large neuron drops are detected with a baseline check, and the Librarian's mission is now explicit — format checking alone doesn't count as a real pass.
If you're running v1, the migration guide walks you through the upgrade in 5 minutes. Your VAULT data is preserved — only system files change.
v1 gave EIDARA memory. v2 made it self-correcting. v3 is where it gets interesting.
We're calling it SUPER DARA — and the idea is simple: your AI shouldn't need you to type everything it should know. It should read, listen, and see.
Reads your conversations and distills them into atomic facts stored locally. Before each new session, the compiler injects the most relevant context into BRAIN.md. Zero manual editing. Less drift, less effort, lower token cost per session.
Listens to your meetings and captures what matters. Not a transcript dump — the specific decisions, action items, and context that make your AI useful in the next conversation.
Sees what you've been doing on your screen. An AI agent starting a new session arrives already aware of your recent work, so you don't have to bring it up to speed.
Everything captured locally, queried locally, never sent to the cloud. Communication happens over MCP — the open protocol — so any compatible AI can plug in.
Same design philosophy EIDARA has had since day one: local-first, file-based, no infrastructure, fully auditable. No new servers. No new accounts. The state of the art keeps getting better — EIDARA absorbs the good parts and keeps the constitution.
Status: research phase. v2.0 is the production-stable release for today.
— Javier