Build a research agent with web search and one bill

A research agent is one of the most useful things you can build, and one of the fastest to set up when search, reading and the model all come through one key. This guide walks through the shape of one.

What it needs

• A model that can plan, read and write a clear answer.
• Search to find candidate sources: Tavily, Exa, Brave Search or Perplexity.
• Reading to turn a promising page into clean text: Firecrawl.
• Reference sources where they help, such as Wikipedia, YouTube or Reddit.

Each of these is reached with your Agent200 API key and billed through Agent200. There are no separate accounts to open and no extra keys to store.

The request

Conceptual example.

const report = await agent200.run({
  model: "openai/gpt-5.6-sol",
  instructions: `
    You are a careful research assistant.
    Search first, then read the most relevant sources.
    Answer in short sections and cite every claim with its URL.
  `,
  input: "What changed in our industry's data regulations this year?",
  tools: ["web_search", "read_page"],
});

What happens inside

1. The model plans and calls the search tool with a few queries.
2. Agent200 runs the searches and returns the results to the model.
3. The model picks the sources worth reading and calls the reading tool for each.
4. Agent200 fetches the pages and returns clean content.
5. The model writes the answer, with citations, and Agent200 returns it to your app.

All of that is one run() request from your side. The loop, the tool calls and the provider requests happen inside Agent200.

Make the answers better

• Ask for citations in the instructions. A model that must cite tends to read before it writes.
• Give it the date. “This year” means nothing to a model that does not know today’s date.
• Pick the model per task. A quick fact check and a long market overview do not need the same model.
• Turn repeat research into a workflow. If you always search, read and summarize the same kinds of sources, a workflow makes it one capability.

One statement

Searches, page reads and model requests all land on the same Agent200 statement. When you compare a cheaper search service with a richer one, or swap the model, it is a change in your request, not a new vendor.

See what it costs on the Pricing page, and every search and data service in the integrations catalog.