Tutorial · Getting started
How to build your first AI agent
In this tutorial you will build a working AI agent from a single plain-language sentence, teach it your documents, and deploy it to a channel — without writing code. Open the sandbox at /try — no account, no card. Type one sentence describing the job, and VegaDūta assembles the agent: tools, guardrails, and a persona you can test immediately.
The free tier runs on the platform-hosted model pool, so there is nothing to configure before your first conversation. When you want a specific model, you can bring your own key from any of 13 providers.
Step by step
Open the sandbox or sign up
Go to /try to build without an account, or sign up at app.vegaduta.ai if you want the agent to persist in your workspace. Everything below works the same in both places.
Describe the job in plain language
Type one sentence: who the agent is and what it does. 'A support agent for my tailoring shop that answers questions about orders and alterations' is plenty. The builder reads it and configures the agent — persona, tools, and sensible default guardrails.
Pick a model
On the free tier the agent runs on the platform-hosted model pool with nothing to set up. If you prefer a specific provider, open Providers and add your own API key — 13 providers are supported, from Anthropic and OpenAI to Ollama and Sarvam.
Add knowledge documents
Upload the documents the agent should know — price lists, policies, FAQs. VegaDūta indexes them for retrieval-augmented generation, so answers are grounded in your material rather than the model's general knowledge.
Test in the playground
Chat with the agent in the playground before anyone else does. Ask the awkward questions your customers actually ask. Adjust the instructions and re-test until the answers hold up.
Deploy to a channel or share a link
Connect the agent to WhatsApp, Telegram, or another channel from the console — or share it with a public link so anyone can talk to it in the browser. Guardrails follow the agent to every channel.
Going further
The one-sentence build is a starting point, not a ceiling.
- Give the agent tools via MCP servers to let it act, not just answer.
- Chain it into multi-step workflows with conditions and human approvals.
- Add voice so the same agent can take phone or in-app calls.
Frequently asked questions
Can I build an AI agent without writing code?
Yes. You describe the job in one plain-language sentence and VegaDūta configures the agent — persona, tools, and guardrails. Refinement happens through the console, not code, and the sandbox at /try lets you do all of this without even creating an account.
Do I need my own OpenAI or Anthropic API key to start?
No. The free tier runs on the platform-hosted model pool, so your first agent works with zero provider setup. When you want a specific model, you can bring your own key from any of 13 supported providers.
How do I make the agent answer from my own documents?
Upload them as knowledge documents in the builder. VegaDūta indexes the content for retrieval-augmented generation, so the agent grounds its answers in your material. You can add and remove documents at any time.
How do people actually talk to the agent once it's built?
Deploy it to a channel — WhatsApp, Telegram, and others — or share a public link that opens a chat in the browser. The same agent, knowledge, and guardrails apply wherever it runs.
See it working in two minutes
The sandbox provisions a real tenant — describe an agent in one sentence and test it, no account, no card. Or browse ~90 industry workflow recipes to see what teams build.
