Building Your First Agentic AI App with SambaCloud and n8n

Agentic AI is a major trend in the evolution of LLMs. Agents enable AI applications to perform more autonomous and context-aware tasks, reducing manual effort.

In this demo, we use SambaNova’s very own SambaCloud and the low-code framework n8n to help beginners easily build a simple agentic application for household budgeting. You’ll learn the basic steps to develop an agentic AI app and gain an intuitive understanding of its core ideas.

Why Use SambaNova for Building Agentic AI Applications?

Our platform offers ultra-fast inference.

For example, the latest Llama 4 model runs at 800 tokens per second on SambaCloud. That’s about 10x faster than conventional GPU-based inference. This speed becomes especially critical in agentic AI applications, such as with deep research agents.

Compared to traditional chatbots or RAG systems, agentic AI apps invoke LLMs far more frequently during execution, as shown in the figure:

Instead of calling just one or two models like a chatbot, an agentic app may call 5-10 different models as part of its reasoning process.

In addition, the number of tokens processed per task increases dramatically, from around 1,000 tokens in a simple chat to 50,000-100,000 tokens in agentic workflows. Naturally, this results in much heavier compute loads and longer processing times if the infrastructure can’t keep up.

By leveraging SambaNova’s ultra-fast inference in SambaCloud, you can build practical and high-UX agentic apps even when the LLM is invoked intensively throughout the workflow.

Agentic AI App Use Case: Household Budgeting

We chose household budgeting because it’s a task familiar to everyone, and it clearly shows how agentic AI can reduce manual cognitive and clerical workload.

The task can be broken down into three main steps:

1. Reading the receipt
2. Structuring the data
3. Recording the data

Technology-Driven Progression

The table below summarizes how the roles of these three steps have shifted with the progression of AI technology.

• Fully-Manual Stage: In the beginning, humans handled all three steps manually: reading receipts, organizing the information, and typing everything into Excel.
• LLM- and VLM-Assisted Stage: Next, we were able to offload part of the process to AI by using technologies like LLMs and VLMs.
  
  • LLMs are particularly strong at cognitive tasks, such as understanding and structuring information. Even then, there were still parts of the process where human involvement was necessary.
• Fully-Agentic Stage: When we move to a fully-agentic AI system, AI is able to carry out not just the thinking but also the actions across all steps.

The difference Between RPA and Fully-Agentic Workflows

One thing to highlight is that automation of actions has been around for quite some time. Tools like RPA have enabled parts of workflows to be automated. However, with conventional RPA, every step had to be manually predefined.

What’s different now is that with agentic AI, the entire workflow, from thinking to acting, can be handled flexibly by AI itself, without the need for rigid, manual definitions.

Three General Approaches to Building Agentic AI Apps

1. Coding everything from scratch: This provides maximum flexibility but requires deep technical expertise and large development efforts.
2. Agentic AI libraries (e.g., LangGraph): These libraries offer frameworks to simplify multi-agent development but still demand strong engineering skills.
   1. The field is also rapidly evolving, with frequent updates to libraries, and no single de facto standard established yet. Even when using libraries, continuous technical catch-up is necessary.
3. Low-code platforms: These platforms allow users to visually design workflows, making it much easier to build agentic AI apps without deep programming knowledge.

Why Use n8n?

In this demo, we chose the low-code platform n8n for three main reasons:

1. Easy to set up: n8n is available as a SaaS service, so users can start using it easily, without complex setup.
2. Simple deployment: For simple chat applications, it can be deployed entirely within n8n itself, without requiring a separate front-end server.
3. Open-source option: n8n also offers an open-source version that can be deployed on-premises, making it a flexible option for enterprise use cases.

How to Use n8n and SambaCloud to Build an Agentic AI Application

n8n allows you to visually design workflows by connecting nodes, with each node representing a step in the process. n8n also has built-in AI agent nodes, which are specifically designed for agentic AI applications. These nodes simplify managing actions, memory, and multi-turn interactions without the need to manually handle complex logic.

In addition, SambaCloud provides an OpenAI-compatible API, so you can connect directly SambaNova APIs to n8n without any special customization.

Household Budgeting Workflow Built with n8n

This flow represents an agentic household budgeting app:

1. The flow starts when a user sends a receipt image or text via chat.
2. The AI agent node processes the input, structures the receipt data into a table format, and asks the user to confirm the extracted information.
3. Once confirmed, the data is reformatted into JSON and then logged to a Google Sheet as the final output.

All of these steps are coordinated flexibly, through prompt-driven interactions, without needing to hard-code every path manually.

Example Interaction with the Workflow

With n8n, once you create a chat workflow, you can open a chat application like this directly from your browser:

First, let’s enter the receipt information into the chat. You can casually describe the receipt in natural language; no strict format or keywords are needed:

The agent responded and noticed that the amount was missing. It flexibly asked for the missing detail. This highlights the flexible, context-aware interaction that is characteristic of agentic AI.

Let’s go ahead and provide the missing information:

The agent processed the input and has structured the information neatly into a formatted entry. It now asks for confirmation.

After confirming, the agent also logs the entry.

As you can see below, the receipt information entered (purchase date, item, amount, store, and category) has been accurately logged into the correct columns of the spreadsheet:

This highlights how an agentic AI app delivers flexible context-aware interaction through natural language. It’s quite convenient, isn’t it?

Ideas for Future Extensions

This demo was intentionally kept simple to focus on the basics. But it can easily be extended to support more complex, real-world workflows. For example, we could add voice input, allowing users to verbally input receipt information.

We could also implement spending analysis via chat, helping users better understand their financial patterns. Or, we could integrate the app with financial service APIs or credit card data for even broader automation.

And of course, there are many more possibilities beyond these. What kind of extensions would you build?

Summary

1. Agentic AI applications can automate both thinking and actions that previously required manual effort by leveraging the flexibility of LLMs.
2. With tools like n8n, agentic AI apps can now be built easily, even by non-developers, without technical barriers.
3. SambaCloud’s ultra-fast inference enables powerful, real-world agentic AI applications.

We hope this demo has sparked your interest in building agentic AI applications. 

If you’re interested, feel free to visit cloud.sambanova.ai and sign up for SambaCloud. Signing up is easy—if you have a Google account, you can get started instantly.

We would love to see you create your own agentic AI applications. We’re excited to see what you’ll build.