Skip to content

ExplanationΒΆ

For anyone who wants to understand Agent Kourier rather than operate it: these pages explain how it is built, why it is built that way, and what it deliberately does not do.

  • Architecture

    What the broker owns, what the agent runtime owns, and the path one conversation takes.

  • Why a broker

    Why a separate, always-on process sits between the chat, the alerts and the agents.

  • Why A2A

    Why one protocol to every agent, what A2A gives the broker, and what it does not.

  • Threads and sessions

    How a thread becomes an agent session, and what happens to replies, restarts and expiry.

  • Delivery guarantees

    What Agent Kourier promises about messages in and posts out, and the windows it accepts.

  • High availability and failover

    Two replicas on Postgres: the Lease, epoch fencing, what a takeover recovers and the duplicates it can leave.

  • Security model

    One identity per Binding, the channel as the access boundary, and why kagent needs a front door.