Research agent template

Q: What does the research agent template build?

A guided builder that configures a research agent: you pick the topic, depth, sources and output format, and the template generates the agent configuration and system prompt.

Build a diligent investigator: pick its beat, shape how it reports back, and draw its guardrails. Four steps, about two minutes.

Documentation

What you're configuring

Topic & depth — the agent's beat. A focused topic ("breakthroughs in fusion energy") produces far better research than a vague one ("energy").
Sources — which source types the agent may consult when you deploy it with tools connected.
Report format — the shape of the deliverable: executive summary, bullet brief, or detailed report.
Quality checks — citations and uncertainty flags keep the output honest.
Guardrails — blocked topics, approval requirements, and a cap on sources per run. These become enforceable rules in the guardrail gate.

What you get

A generated system prompt ready to paste into any agent framework, plus the full configuration JSON — which will flow straight into the guardrail gate when it launches.

Tips

Require citations if anyone else will read the output.
Block topics you never want it near (medical or financial advice, for example).
Keep "max sources" modest — more sources isn't always better research.

FAQ

What does the research agent template build?

A complete agent configuration: topic, depth, sources, output format, quality checks and guardrails — expressed as a system prompt and JSON you can copy.

Does it browse the live web?

That depends on the provider and tools you connect when you deploy with your own API key. The template lets you declare which sources the agent may use.

How are results checked?

Through the guardrail gate: output checking reviews results before they reach you, and you can require human approval before anything is published or sent.