Build your first agent
From signing in to a working agent. This walkthrough builds a small news agent with a generated chat UI — about twenty minutes, and every concept you will use later appears in it once.
1. Sign in and create a project
- Go to flux.conceptune.com and sign in with your Conceptune account.
- You land in the Console.
- Create a project if you do not have one — it is the container for modules, credentials and billing.
2. Create a module
- From the sidebar, select Agentic AI.
- Click Add to create a new module.
- Fill in the Name and Description.
- Click Create Module.
Your module appears in the list. Hover it and a Workflow button appears — click it to open the playground.
The Agentic AI module list with one module card, showing the Workflow button revealed on hover
Screenshot to be added3. Add a trigger
Nothing runs without a trigger. For a first agent, the UI trigger is the easiest — Flux generates a chat interface for you, so you can talk to the agent without connecting a channel first.
- Open the Nodes panel on the left of the playground.
- Drag the UI trigger onto the canvas.
- Click it to open its form and give it a name.
4. Add an AI node
- Drag a GPT Model node — or Gemini Model — onto the canvas.
- Wire the trigger's output to the AI node's input by clicking one connector and then the other.
- Open the node and write its system prompt: what the agent is and how it should behave.
- Pick a credential. Leave it on the default to use Flux's own API keys — usage is billed to your project — or select your own from the credential store.
In the node's input, reference what the trigger produced using the template language. You do not have to type it — drag the key from the connected node's Response JSON Keys list straight into the field.
5. Add a response
- Drag a Chat response node onto the canvas.
- Wire the AI node's output into it.
- In its message field, reference the model's answer with
{{POUT ...}}.
You now have the smallest complete workflow: trigger → AI → response.
A three-node workflow on the canvas: UI trigger wired to a GPT Model node wired to a Chat response node
Screenshot to be added6. Give the agent a tool
This is the step that makes it agentic rather than a single prompt. A tool is something the model may choose to call.
- Drag an API tool node onto the canvas.
- Attach it to the AI node's tool connector rather than its input.
- Configure the request — for a news agent, a headlines endpoint.
- Describe each parameter with the
DESCkeyword so the model knows what to put in it:{{DESC 'The topic to search news for'}}.
The model now decides for itself whether a question needs the news API, and fills in the parameters from what the user asked.
7. Test it
Open the generated UI from the trigger node and talk to your agent. Then read what actually happened in chat logs and compiled logs — the compiled log shows each template after it was resolved, which is where most first-attempt problems reveal themselves.
8. Go live
When it behaves, open the module and switch on Production mode. See Going to production for what to check first.
What to read next
- Nodes and their types — the full palette.
- Node Template Language — moving data between nodes.
- Messaging channels — putting the agent on WhatsApp or Telegram.