mikrokhoros
Open source AI agent harness for object-first worlds
2 followers
Open source AI agent harness for object-first worlds
2 followers
Open source AI agent harness for object-first worlds. Every persistent world capability an agent can invoke is an exact concrete object with declared functions, identity, state, location, authority, and history; the runtime enforces locality, possession, permissions, bounded actions, and replayable trajectories.
Hi Product Hunt — I built mikrokhoros because I wanted agents to live inside a world, not just call disconnected tools.
In mikrokhoros, persistent capabilities are exact objects. Each has identity, state, location, authority, history, and declared functions. The agent observes a bounded situation and proposes a small action; the runtime decides whether it is valid, commits the consequence, and carries authoritative state into the next observation.
That gives you replayable trajectories where success, failure, possession, locality, and permissions are part of the environment rather than prose in a prompt.
It is open source under Apache-2.0, built in Swift 6.1, and free to use. The project is pre-release. I would especially value feedback on the default khoros and the object interfaces people would want to build for agents.