Two live agents sharing memory across two nodes.
The agents are ready to chat and remember. They run on separate polign_db nodes that hold nothing locally and share one backend store, so the only place that memory exists is the bucket.
Sign in to get your own namespace
The demo uses your Polign account only to give you a memory space nobody else can read. The nodes enforce that boundary with a key bound to your namespace. No account yet? Signing up takes a minute with an email or a Google account.
What is happening
Every line in the terminals is a real tool call against a real database. Nothing is scripted.
Your own namespace, enforced by the server
Signing in mints a polign_db API key bound to a namespace derived from your account. The node stamps every write with it and constrains every read to it, so the key cannot reach anyone else's records no matter what the agent sends.
Two nodes, one store, nothing in between
Agent A talks to one cold-first node, agent B to another. Both serve reads from the same bucket prefix and the shared write log. A memory crossing from A to B never passed through a process; it went through the store, and a write token makes the second node wait until it has caught up.
Typed memory with history
The agents use Recall, an open memory library: a closed set of predicates, typed values, and corrections that replace a belief without erasing what it replaced. Ask what you said before and you get the record, not a guess.
The Wikipedia questions go to the same read-only index behind demo.polign.com. Read the agent memory at scale post for the numbers behind the cold path, or run polign_db yourself.