Search Data API for AI Agents and LLM Tooling

Key API parameters for agent workflows

Agent pipelines usually need to control freshness, scope, and prompt size more than anything else. OrbitScraper exposes those controls directly on the search request so you can keep the retrieval layer deterministic.

Code example — agent tool definition

A practical pattern is to expose OrbitScraper to the model as a tool, let the model provide q and freshness controls, then map that tool call straight to the search endpoint.

{
  "type": "function",
  "function": {
    "name": "search_web",
    "description": "Search the web and return structured results or a markdown summary for grounding LLM responses with current information.",
    "parameters": {
      "type": "object",
      "properties": {
        "q": {
          "type": "string",
          "description": "The search query"
        },
        "markdown": {
          "type": "boolean",
          "description": "Set true to return a prompt-ready markdown summary instead of raw JSON",
          "default": true
        },
        "time_period": {
          "type": "string",
          "enum": ["past_day", "past_week", "past_month", "past_year"],
          "description": "Optional freshness filter"
        }
      },
      "required": ["q"]
    }
  }
}

curl -X POST "https://api.orbitscraper.com/v1/search" \
  -H "x-api-key: ORS_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "q": "latest LLM benchmarks 2026",
    "markdown": true,
    "time_period": "past_week",
    "engine": "google"
  }'

{
  "result": {
    "markdown": "# Search Results\n\n## Latest LLM Benchmarks 2026\n...",
    "organic_results": [
      {
        "position": 1,
        "title": "...",
        "link": "...",
        "snippet": "..."
      }
    ],
    "detected_intent": "informational"
  }
}

RAG pattern

Use OrbitScraper as the retrieval step in a RAG pipeline. Query the API for current documents, take the normalized snippets from organic_results, and inject either those snippets or the markdown block into the LLM context window.

The markdown field is especially useful when you want one prompt-ready context block without writing a custom summarizer for every result set.

1. A user query arrives at your agent or application.
2. The agent calls OrbitScraper with the query and markdown: true, optionally adding engine, time_period, and num controls.
3. The markdown string or selected snippets are injected into the system prompt, tool result, or retrieval layer before the agent answers.

LLM frameworks

OrbitScraper integrates naturally with LangChain as a custom Tool, with LlamaIndex as a retrieval connector, and with any framework that supports function calling or external tools.

• LangChain: wrap POST /v1/search plus GET /v1/search/:jobId as a single tool.
• LlamaIndex: treat normalized organic_results as an external retriever that updates in real time.
• CrewAI and AutoGen: register OrbitScraper as a shared team tool for web context, market scans, or routing decisions.
• Custom stacks: store the structured JSON for auditability, then pass only the markdown or selected snippets into model prompts.

When teams add Research API

SERP API is the right first step when the agent needs raw retrieval context, result modules, and prompt-ready markdown. Teams add Research API when they also want OrbitScraper to fetch supporting pages and return a synthesized answer with source objects attached.

curl -X POST "https://api.orbitscraper.com/v1/research" \
  -H "x-api-key: ORS_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "Which open-source LLM frameworks are gaining enterprise adoption?",
    "depth": 5,
    "output_format": "summary",
    "include_sources": true
  }'