Skip to Content
TutorialsFirst agent (prompts)

Tutorial: First agent (prompts)

Create an agent by asking the Agentron assistant in natural language. No UI forms: just prompts that work.

Prerequisites

  • Agentron running (npm run dev:ui, open http://localhost:3000 )
  • At least one LLM provider configured (e.g. Settings → LLM providers: OpenAI, Anthropic, Ollama, or OpenRouter)

Prompts to try

Open Chat and send these in order (or adapt them).

1. See what you have

  • “What LLM providers do I have?”: The assistant calls list_llm_providers so you know which config to use.
  • “List my tools.”: See built-in and any custom tools (e.g. std-weather, std-fetch-url).

2. Create a simple agent

  • “Create an agent named Hello Agent. It should greet the user. Use my first LLM provider.”

    The assistant will use create_agent with a name, description, and a minimal node graph (e.g. input → LLM → output). It will attach an LLM config from list_llm_providers.

3. Give the agent a tool (optional)

  • “Add the weather tool to my Hello Agent.” (If you have a weather tool, e.g. std-weather.)

  • Or: “List tools, then add std-fetch-url to the Hello Agent.”

    The assistant uses get_agent, list_tools, and update_agent with toolIds.

4. Run or inspect

  • “Run my Hello Agent with input: Say hi.”: The assistant can use the run API or point you to the Runs page.
  • “Show me my Hello Agent.”: get_agent and a summary.

What the assistant did

Behind the scenes, the assistant called tools such as:

  • list_llm_providers: to pick an LLM
  • create_agent: with name, description, llmConfigId, graphNodes, graphEdges
  • list_tools / get_agent / update_agent: to add tools

You can do the same via the Agents UI; the chat is another way in.

Next

Last updated on
DownloadFirst workflow (prompts)