Build First Agent
Build and run your first AI agent in ByteChef: an OpenAI model with one tool, in a few minutes with a single API key.
In this guide you build the same workflow as the Build your first agent template: an AI Agent step with an OpenAI model and one tool, followed by a logger. The agent fetches a random quote through the tool and explains it in one sentence. The only thing to set up is an OpenAI connection.
You will learn:
- How to add an AI Agent step and open the AI Agent Editor
- How to pick a model and create a connection for it
- How to give an agent a tool and tell it when to use it
- How to try an agent in the playbook and read its tool calls
Before you start
- Finish Build First Workflow if you have not. This guide starts where it ends: signed in, with a project.
- Have an OpenAI API key ready.
Create a workflow
On your project's row, click + Workflow, enter the label Build your first agent, and click Save. The editor opens with a Manual trigger in place.
Add the AI Agent
Click the + button below the trigger, search for AI Agent, select it, and choose the Chat action.
The agent step appears on the canvas with a warning: it has no model yet.
Click the agent step. The AI Agent Editor opens: configuration on the left, an Agent Playbook for trying the agent on the right.
Pick the model and connect OpenAI
Click Select a model... and choose OpenAI from the list of providers.
Click the gear icon next to the model to open its settings, then open the Connection tab and click +. In the Create Connection dialog, give the connection a name such as OpenAI, paste your API key into Token, and click Save. The connection is stored once and reused by every step that needs OpenAI.
Open the Properties tab and choose a model, for example gpt-5-mini. Leave the rest at its defaults, then close the settings panel with the X.
Write the prompts
In Instructions to follow, enter the system prompt:
You are a friendly assistant. Always call the Get a random quote tool first. Then reply with the quote, its author, and one sentence explaining what the quote means. Do not invent quotes.In User input, enter the message the agent gets on each run:
Fetch a random quote and explain it.Give the agent a tool
Click Add Tool, type HTTP in the filter, select HTTP Client, and choose its GET action.
Open the tool's ⋮ menu and click Configure. On the Properties tab, set URI to:
https://dummyjson.com/quotes/randomLeave Response Format on JSON. No connection is needed, the API is public. Tool Name and Tool Description can stay empty: they default to the action's own.
On the Description tab, set Title to Get a random quote, then close the panel. The agent sees tool titles, and the system prompt refers to this one by name.
Try it in the playbook
Click Test Agent on the right. The playbook opens a chat with your user input already filled in. Send it. The agent reads the prompts, decides to call the tool, receives the quote, and writes its explanation. Expand the HTTP Client: GET entry above the reply to see why the agent called the tool, what it sent, and the JSON it got back.
Send the same message again and the agent explains a different quote. Close the editor with the X when you are done.
Log the answer and test the workflow
On the canvas, the agent step now shows the OpenAI and HTTP Client icons for its model and tool.
Click the + button below the agent step, search for Logger, choose the Info action, and enter this in the Text field:
${aiAgent_1}This logs the agent's reply.
Click Test at the top right of the editor. The execution panel opens below the canvas. Click the agent step to read its reply on the Output tab.
What you have built
Manual Trigger → AI Agent (OpenAI model + HTTP Client tool) → Logger- Model: the agent's reasoning. Any of the supported providers plugs into the same slot.
- Tools: every component action can be a tool. The agent decides when to call it and fills in its parameters. See AI Agent for tools whose parameters the model supplies at runtime.
- Connection: created once, reused everywhere, and never part of an exported workflow.
The same workflow is available as a template. To import it instead of building it, open the Build your first agent page, click Use template, select your OpenAI connection on the model, and click Test.
Next steps
- Star Repository on GitHub: a workflow with an OAuth connection
- Configure Workflow Trigger: a real trigger, a deployment and monitoring
- Add Memory or Guardrails to the agent from the Advanced view of the AI Agent Editor
How is this guide?
Last updated on