Waku Agent tutorial: loops, graphs and dashboard setup

Sean Chen explains loops and graphs through Waku Agent, which the supplied description calls an open source, local-first assistant. His distinction is practical: a loop lets the model choose its next tool call, while a graph defines the workflow's shape in advance. He argues that graphs can contain loops, especially for exploratory tasks such as web research.

The tutorial connects prompt engineering, context and skills to the surrounding agent harness. It then covers installation with pip install waku-agent, environment keys and launching a local dashboard. The demonstration uses calendar services, GitHub and web search; it does not establish that the whole workflow runs offline or uses a local LLM.

Two workflows illustrate the design. Triage classifies a request while checking the calendar, then can hand work to an agent loop. The /gather command retrieves GitHub, web, calendar and memory context in parallel before synthesizing a response. Chen compares this with his sequential loop example without presenting measured performance benchmarks.

The code walkthrough shows graph nodes, edges and Python workflow files. Chen explains how contributors can add workflows and closes with his view that model calls and model-selected routing distinguish these graphs from fully deterministic scheduling.