Hey Product Hunt 👋 I'm Ansgar, one of the founders of Tiamat Labs. We didn't build Kepos to launch a product.We built it because we were drowning in our own mess. Between running paid media accounts for a handful of clients and building out several product ideas in parallel, everybody had Claude, Cursor, Codex, and half a dozen agent sessions running at once, and every single one of them forgot everything the second the session ended. I'd re-explain the same campaign structure, the same client quirks, the same "no, we tried that in March and it didn't work" three times a week. My team had the same problem from the other side. Knowledge scattered across docs, Slack, and whoever happened to remember it. So we built the thing we actually needed: one memory layer that both our AI agents and our team read from and write to. Git-backed, so nothing silently overwrites anything and we can see exactly what changed. And the part I care about most: when we correct an agent once, that correction gets recorded and enforced going forward, instead of the same mistake showing up again in the next session. We've been running on it internally for a while now across real client work. Not a demo, our actual day-to-day. Today's the first time it's public. I'll be around all day. Happy to answer anything. Architecture, how the multi-agent piece actually works, what we got wrong the first three times we tried to build this. Fire away. Ansgar

Kepos

Shared memory for teams and their AI agents

Shared memory for teams and their AI agents