Skip to content

The kaboo stack

kaboo is a small stack for building multi-agent apps with live, inspectable agent activity: define agents in YAML and serve them as AG-UI SSE (kaboo-workflows), persist and replay the full event log inside a CopilotKit runtime (kaboo-runtime), and render hierarchical agent activity with drill-down and human-in-the-loop in React (kaboo-react). The kaboo-workflows-demo assembles all three into a runnable reference.

flowchart LR
  subgraph Python
    W[kaboo-workflows<br/>YAML agents → AG-UI SSE]
  end
  subgraph Node
    R[kaboo-runtime<br/>CopilotKit AgentRunner + ThreadStore]
  end
  subgraph Browser
    C[kaboo-react<br/>agent-activity UI]
  end
  DB[(Postgres)]
  C -->|/api/copilotkit| R
  R -->|HttpAgent /invocations| W
  R <-->|persist / replay| DB
  W -->|AG-UI events| R
  R -->|streamed events| C

New here? Start here to pick a path, or follow the full-stack tutorial to wire all three together.

Libraries

  • kaboo-workflows — YAML multi-agent orchestration → AG-UI SSE (Python / PyPI).

    Docs · Repo

  • kaboo-runtime — CopilotKit runtime persistence: a custom AgentRunner + pluggable ThreadStore that replays the full event log.

    Docs · Repo

  • kaboo-react — React components + hooks for live, hierarchical agent activity inside a CopilotKit chat.

    Docs · Repo

  • kaboo-workflows-demo — a runnable, end-to-end market-research demo wiring the three libraries together, shipped as an example in this repo.

    Example

Run the demo

The fastest way to see the whole stack in action is the demo — a market-research assistant with delegated sub-agents, human-in-the-loop approvals, and full replay across reloads.

  1. Clone this repo and open the examples/kaboo-workflows-demo directory.
  2. Follow the validated startup: Postgres via Docker, then the pipeline (:8080), backend (:4000), and frontend (:3000).
  3. Open http://localhost:3000 and ask it to research a market.

Contributing

One shared model spans every repo — see Contributing to the kaboo stack.