Now Available

Search API for
AI Agents

Access an AI-powered index of 25B+ pages through a single API. Auto-selected search modes, structured JSON responses, and output formats built for LLMs.

Automatic mode selectionSub-second fast responseRAG-ready output

Using Claude Code, Cursor, or Codex? Paste this line:

agent

Set up Andi search for this project: andiai.com

Your agent reads install.md and configures everything.

Get API access Read the Docs

Already have access?Open Console·Agent setup

terminal

$ curl -s "https://api.andiai.com/api/v1/search?q=best+hiking+trails" \
  -H "x-api-key: your_key" | jq .

{
  "results": [
    {
      "title": "Best Hiking Trails in the US",
      "link": "https://example.com/trails",
      "desc": "From Zion Narrows to...",
      "source": "example.com"
    },
    // ... more results
  ],
  "results_type": "search"
}

25B+

Pages indexed

Independent search benchmarks

<1¢

Per query

<1s

Fast mode response

Hands-Free Search, Full Control

Andi determines what each search needs — how much compute, which models, how deep to read — and adjusts automatically. Set the parameters yourself only when you want to.

Auto — the default

Send the query and nothing else. Simple lookups come back in under a second; hard questions automatically get deeper retrieval and more thorough reading. No tuning, no mode-picking logic in your agent.

searchMode=auto · default · adapts per query

Manual control when you want it

Pin an effort level with searchMode when you want the same behavior on every call — low-cost, fast (sub-second), balanced, deep (~2–3s, thorough), or exhaustive (maximum coverage). Full reference in the docs.

Built for AI Agents & Developers

Every surface — API responses, documentation, this website — is built for programmatic consumption. Whether you're an AI agent, a coding tool, or a developer with curl.

Fetch Endpoint

GET /api/v1/fetch?url=... reads any web page as clean extracted content — JSON or LLM-ready markdown. Failed fetches are free.

Outcome-Based Pricing

Priced by what it took to find and get the results — easy lookups cost less, hard-to-find content costs more. Every JSON response reports the exact charge in metrics.cost_dollars.

MCP Server

Connect Claude Code, Cursor, or VS Code to api.andiai.com/mcp — a remote MCP server with web search and URL fetch tools. Streamable HTTP, authenticated with your API key. Claude Code can also install it as a plugin — skill and MCP server in one step.

LLM-Optimized Output

format=context returns search results as markdown built for LLM context windows. Feed directly into RAG pipelines and AI agents.

One-Call Content

Searches with format=context return ranked results with query-related extracts — the passages from each page that answer the query, attributed to their sources. Add content=true for full page content. One request replaces the search-then-fetch-then-summarize loop.

Multi-Query Search

Pass a JSON array of up to 5 queries in q and get merged, deduplicated results in one response. A research fan-out costs one round trip.

Accurate Results

Built on Andi's own index of 25 billion pages, with every page scored for meaning and credibility at ingestion. 87% accuracy in independent search benchmarks.

Standards Compatible

Google CSE and OpenSearch-compatible /api/v1/opensearch endpoint as well as our AI-friendly /api/v1/search. Migrate from an existing search API with minimal code changes.

Get Started

Get your API key

Read the docs

Quickstart, parameters, response formats, and code examples at docs.andiai.com.

Make your first query

One curl command to search 25 billion pages.

quickstart

$ curl "https://api.andiai.com/api/v1/search?q=your+query" \
  -H "x-api-key: YOUR_API_KEY"

Start Building

Get access to 25 billion pages of web content through a single API. Auto picks the right search mode for each query, structured for AI agents.