# Quickstart

Build and test your first agent in 5 minutes. By the end you'll have a working "hello world" workflow with one LLM agent that responds in a chat preview.

## 1. Sign in

Open the app and sign in with your tenant credentials. (New accounts need an administrator's approval before the first login — see [Billing & account](#/docs/billing).) The left sidebar has six entries:

- **Workflows** — your agents, in a table with status dots and run/token stats
- **Production** — the back office of each deployed agent
- **Credentials** — encrypted tokens for APIs and integrations
- **Marketplace** — installable templates
- **Billing** — credit balance and usage
- **Documentation** — what you're reading now

> *Screenshot: left sidebar with Workflows / Production / Credentials / Marketplace / Billing / Documentation entries.*

## 2. Create a new workflow

On the **Workflows** page, click the floating **+** button (bottom right). The New Workflow modal offers three starts: an **empty workflow**, a **template**, or an Agent Builder JSON import. Pick empty, give it a name (e.g. `hello-agent`) and confirm. You'll land on the editor with an empty canvas — except for a single **Start** node already placed for you.

> *Screenshot: empty editor canvas with a Start node in the middle.*

The left panel is the **Node library**. Categories shown:

```text
Core          Agent · Classify · End · Note
Tools         Guardrails · MCP
Triggers      Webhook · HTTP Request · Respond · Schedule · Event listener
Integrations  Send Email
Logic         If / else · While · User approval
Data          Transform · Set state · Generate file · RAG
```

## 3. Add an Agent node

Drag **Agent** from the Core category onto the canvas, next to the Start node. Click the small handle on Start and drag it onto the Agent node to connect them. You should see a curved edge:

```text
[Start] ────▶ [Agent]
```

> *Screenshot: Start node connected to an Agent node by an edge.*

## 4. Configure the agent

Click the Agent node. The right side opens the **Config panel**. Fill in the basics:

| Field | Value |
|-------|-------|
| Name | `Greeter` |
| Model | `gpt-4.1-mini` (or any model your tenant has enabled) |
| System prompt | `You're a helpful assistant. Keep answers short.` |
| Output mode | `text` |

Leave tools empty for now. The Output mode controls how downstream nodes see the result:

- `text` — raw string
- `json` — parsed JSON object
- `structured` — typed schema (you define the fields)

## 5. Test it in the chat preview

Click the **Chat** button (top right of the editor) to open the preview pane. Type a message:

```text
hello, what can you do?
```

The Greeter agent runs and streams its reply. You'll also see the **Run** appear in the run history with timing, token usage, and the full trace of node executions.

> *Screenshot: chat preview on the right showing a user message and an assistant reply, with a run entry in the history.*

> Your new workflow starts as **Draft** — the chat preview runs it regardless, but to serve real traffic you must switch it to **Live** from the Production space. Going Live runs a readiness check first (it won't activate if, say, an Agent has no instructions or an HTTP node has no URL).

## Next steps

- Add an **HTTP Request** node to call an external API
- Plug in a **Function tool** for custom logic
- Wire a **Webhook** trigger to start the workflow from outside
- Activate the agent and expose it from the **Production** space (Deploy → API / Widget / Shared chat)
- Browse the **Marketplace** for ready-made templates

---

*Source: https://agentbuilder.systalink.sn/docs/quickstart — human documentation.*
*Other language: [/docs-md/fr/quickstart.md](/docs-md/fr/quickstart.md).*
*Machine-readable index: [/llms.txt](/llms.txt).*