Skip to main content

Building a Research Team

In this guide you'll build a team that researches a topic and writes a structured report. It has three specialists:

  1. Research Planner: breaks a topic into research questions.
  2. Web Researcher: searches the web and gathers evidence.
  3. Report Writer: turns the findings into a report.

You'll need at least one model under Resources → Models, and a web search function under Resources → Functions (or one from the Marketplace).

Step 1: Create the agents​

Go to Build → Agents and click +. For each agent, fill in Agent Name, System Message and LLM Configuration, then click Create Agent.

Research Planner​

Default Role: Breaks a topic into research questions

You are a research planning specialist. When given a topic, break it down
into 3-5 specific research questions that together give comprehensive
coverage of the topic.

Output: a numbered list of questions, each with one line on why it matters.

Web Researcher​

Default Role: Finds and cites evidence on the web

You are a web research specialist. For each research question you are given,
search for recent, credible sources.

For each finding include the source URL, the key facts, and how relevant it
is. Prefer recent sources and cross-check claims where you can.

Report Writer​

Default Role: Writes the final report

You are a research report writer. Turn the research findings in this
conversation into a clear report with: an executive summary, key findings
grouped by theme, evidence with source citations, and a conclusion.

Write for a professional audience. Be concise but thorough.
tip

Default Role is how each agent is described to its teammates. It matters most in Group Chat Selector teams, but it's worth setting on every agent.

Step 2: Give the Web Researcher its tools​

  1. Open Web Researcher and go to its Configure tab.
  2. Click Assign Functions, search for your web search and page-fetch functions, select them, and click Assign.

The planner and the writer don't need tools. See Adding Tools to Agents.

Step 3: Test each agent on its own​

Click Open in Chat on each agent and give it a typical task. Adjust its system message until the output looks right. This is much easier to do before the agents are on a team.

Step 4: Create the team​

  1. Go to Build → Teams and click +.
  2. Enter Team Name: Research Team.
  3. Set Team Type to Sequential. The planner runs first, then the researcher, then the writer.
  4. Click Assign Agents and add the agents in run order: Research Planner, then Web Researcher, then Report Writer.
  5. Leave the pre-filled routing fields as they are. A Sequential team runs its members in the order you added them.
  6. Leave Interaction Mode as Default (Supervised).
  7. Click Create Team.
tip

Sequential suits this team because each member builds on the previous member's output. If you'd rather have the members discuss and refine the report together, use Group Chat. If the order of work depends on the topic, use Group Chat Selector. See Team Orchestration Patterns.

Step 5: Chat with the team​

  1. On the team's page, click Open in Chat.
  2. Send a topic, for example: "What are the main approaches to AI agent evaluation in 2026?"
  3. Watch the planner list the questions, the researcher gather sources, and the writer produce the report.

Step 6: Add regression tests​

  1. On the team's Tests tab, click Add Test Case.
  2. Enter a representative topic as the Input Prompt and a short Expected Output describing what a good report contains.
  3. Choose the metrics. GEval with a Rubric such as "Has an executive summary and cites a source for every finding" works well for reports.
  4. Click Run Test after every change to the agents.

See Testing Agents and Teams.

Step 7: Use the team in a workflow​

  1. Create a workflow under Build → Workflows.
  2. Add a Team node and select Research Team.
  3. Feed the topic in from a workflow attribute.

You can then run the workflow on demand, on a schedule, or from a webhook. See Designing Workflows and Workflow Attributes.

Next steps​

  • Add specialists, such as a data analyst with database tools.
  • Add an approval gate before the report is published. See Using Approval Gates.
  • Check token use and cost on the team's Observability tab. See Observability.