Tutorial: First workflow (prompts)

Create a workflow that connects two agents by asking the Agentron assistant. No graph editor required: just prompts.

Prerequisites

• Agentron running and at least one LLM provider configured
• Either two existing agents, or be ready to create them via chat (see First agent)

Prompts to try

Open Chat and send these (or adapt).

1. Create two agents (if you don’t have them)

• “Create an agent named Researcher that uses the first LLM provider and can fetch URLs.”

• “Create an agent named Summarizer that uses the first LLM provider and summarizes text.”

  The assistant uses create_agent (and optionally list_tools / update_agent for tools like std-fetch-url).

2. Create the workflow

• “Create a workflow named Research then Summarize. Add the Researcher agent and the Summarizer agent, and connect them in that order. Set max rounds to 2.”

  The assistant will:

  • create_workflow with a name
  • update_workflow with nodes (two agent nodes) and edges (Researcher → Summarizer)
  • set maxRounds so the run doesn’t loop forever

3. Run the workflow

• “Run the Research then Summarize workflow with input: Summarize the content of https://example.com .”

  The assistant can trigger execution (e.g. via execute_workflow or by pointing you to the workflow’s Run button). You can check Runs for output and logs.

4. Inspect or fix

• “Show me my Research then Summarize workflow.”: get_workflow
• “The last run failed. What went wrong?”: The assistant uses list_runs, get_run, and optionally get_workflow / get_agent to diagnose (see AI Guide: diagnose before fix).

What the assistant did

• create_workflow: empty workflow with a name
• update_workflow: nodes (type agent, config.agentId), edges (source → target), maxRounds
• execute_workflow: run with input

Next

• First RAG (prompts): Add a knowledge collection and use it in chat or an agent
• Concepts: Workflows: Execution modes, branches, scheduling