Concepts Overview

Meerkat is easiest to understand as a composable agent runtime with a practical product layer on top. You can approach it from any surface:

The core concepts fit together like this:

Read order

If you are new to Meerkat, a good concepts path is:

Then branch into the systems you need:

Concept	Main question
`Composability`	What kind of product is Meerkat, and why do the surfaces behave like peers?
`Realms`	What defines shared vs isolated state?
`Sessions`	How does one conversation live, continue, and end?
`Configuration`	How are runtime settings stored and updated?
`Providers`	How are models resolved, selected, and capability-gated?
`Tools`	How does the agent act on the world?
`Auth & bindings`	Under what identity/credentials does the runtime talk to a provider?
`Memory & compaction`	How does Meerkat handle context beyond one model window?
`Comms`	How do long-lived agents exchange messages?
`Mobs`	How does Meerkat coordinate multiple agents as one system?
`Scheduling`	How does durable automation fire over time?
`Live Channels`	How does low-latency interactive transport fit into the session model?

After the core concepts, choose the path that matches the system you are building:

Path	Start with
Embedding Meerkat in Rust	Rust SDK overview
Driving Meerkat from an app	Python SDK or TypeScript SDK
Running Meerkat as a service	JSON-RPC or REST
Building multi-agent systems	Mobs
Running private models	Self-hosting models
Understanding ownership	Architecture