Technical

Any AI can now feed my memory. None of them touch it.

Published May 30, 2026

The old model

Before: every AI that wanted to contribute to my memory had to open the DARA folder, read the constitution, and write directly to the vault. It worked, but it was fragile. Two AIs writing to the same file at the same time? Concurrency problems. An AI from a platform with unreliable tool calls making a bad edit? Data corruption. The governance was in the rules, but the enforcement was honor-based.

The new model: many producers, one curator

Now there's a single channel. Any AI — Claude, GPT, DeepSeek, whatever — can contribute an atomic fact via MCP, the open protocol. "Javier decided X on this date." "Project Y moved to phase Z." One self-contained assertion per call. The fact lands in a transit zone. It doesn't touch the curated brain.

What happens in the transit zone

A local LLM running every 15 minutes picks up new facts and runs them through a pipeline. First, a SHA-256 hash check kills exact duplicates without burning any inference. Then a keyword-plus-entity search finds the 5 most relevant existing memories. Then the LLM makes a decision: store it in the right neuron, skip it if it's already known, update an existing fact if the new one is more recent, or merge two related entries. If there's a genuine conflict that the automated rules can't resolve — rare — it escalates to a human queue.

The death of "open the folder"

DARA is no longer a folder you navigate. It's a service that any AI queries via protocol. The MCP enforces the constitution by construction — external AIs physically can't write to the curated brain. They can only contribute to the transit zone and let the autonomous curator do its job. The security isn't in telling AIs "please follow the rules." It's in making it impossible not to.

The result

I now have Claude, GPT, and DeepSeek sessions all contributing facts to the same persistent memory. None of them interfere with each other. The curator processes everything, every 15 minutes, and the next AI session starts with all of that knowledge already integrated. No manual syncing. No copy-pasting context between platforms. One memory, many contributors, one curator.

— Javier

---

See the full roadmap →