Product
Cosmos
Agents that decide as you would.
Cosmos stops your agents from doing things you would not. Lab surfaces write into one account so recognition compounds.
The bet
Teams are shipping agents that can act. Almost none decide as you would.
Cosmos bets that the scarce layer is judgment: a signature agents call before they act, so they hold what matters and suppress what does not.
What it does
Agents call Cosmos before they act. It returns recommend, refuse, or escalate, with a why.
We do not prescribe which tools the agent uses. The agent works. Cosmos is the judgment call. Every hold, refuse, accept, and correct deepens the signature.
What we are learning
Built off how memory is stored in the brain. Restuffing a context window when you say remember is not memory.
Two provisional patents sit under that work: neuroanatomically grounded knowledge graphs, and conversational connectomics. The product is live attention orchestration.
Inspectable narrative
Every decide comes with a receipt: call, return, why, and the evidence that earned the why.
Polarity Lab runs on Cosmos daily across the stack. Surfaces like Lockbox, AVDP, WaxFeed, GPS, and Blueno train the same signature.
Where this goes
Sold to agent teams that need those agents to act like them. Hard domains are where the signature gets stress-tested.
Try the product at cosmos.polarity-lab.com. Depth and demos live under How it works on this site.
Roles we are looking for
Platform engineers
Graph, sync, imports, and hosted infrastructure.
MCP & integrations
Connectors, client compatibility, developer experience.
Alignment & readouts
Turn graph signal into narrative people can trust and use.
Research collaborators
Studies on narrative, suppression, and whether reflection changes behavior.