Kagi Search API

AI model · Reranker / retrieval · Text → Tables / structured data by Kagi

Better results in. Better answers out. Meta-search API for AI agents.

A high-recall meta-search API for AI agents that aggregates 12 sources in parallel. It features AI slop filtering, privacy proxying, and clean Markdown extraction to provide verifiable data for LLMs.

Model specs

The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.

Languages
English

Category: AI Infrastructure

License
Licensed for commercial and AI use
Upstream Sources
12 providers per query
URL Recall Rate
41% (Benchmark leader)
Token Efficiency
35% fewer tokens per correct answer
Extract Limit
Up to 10 URLs per request
Response Formats
JSON, Markdown
  • Uses 35% fewer tokens per correct answer in AI agent workflows
  • Fans out every query to 12 sources in parallel for maximum recall
  • Licensed for commercial and AI use with built-in privacy proxying

Overview

Kagi Search API is a high-performance meta-search API designed specifically for AI agents and developers who require verifiable, high-quality data. It solves the problem of "AI slop" and search blind spots by fanning out every query to 12 different search providers in parallel and re-ranking the results in a single pass. The service includes SlopStop technology, which uses community reporting and auto-classification to downrank AI-generated content in favor of original human work. Every query is proxied through Kagi to ensure that upstream providers never see the identity of the user or agent, maintaining strict privacy. It is licensed for commercial and AI use and provides official SDKs for Go, Python, TypeScript, and Rust.

Search and Ranking Quality

  • 12-Source Parallel Search: The API fans out each query to 12 search providers in parallel to ensure gaps in one index do not become gaps in results.
  • Intelligent Re-ranking: A re-ranking pass is performed on every query to ensure the most relevant and trustworthy sources rise to the top.
  • AI Slop Filtering: SlopStop technology flags and downranks AI-generated "slop" pages across web, image, and video results based on community reporting and auto-classification.
  • Privacy Proxying: Every query is proxied through Kagi so that upstream search providers never see the identity of the user or the AI agent.

Data Retrieval and Extraction

  • Multi-modal Search Results: The Search endpoint provides real-time results across web, images, news, videos, and podcasts.
  • Clean Content Extraction: The Extract endpoint returns clean page content as readable text or Markdown, stripped of navigation, ads, and clutter.
  • Granular Search Filters: Users can scope results using specialized lenses, country filters, and language settings.
  • Domain Personalization: Results can be personalized with domain ranking and custom URL rules.

Developer Tools and Integration

  • MCP Server Integration: Developers can plug into a hosted Model Context Protocol (MCP) server at mcp.kagi.com/mcp.
  • Commercial Licensing: The API is licensed for commercial and AI use, making it suitable for research and compliance-heavy applications.
  • Interactive Playground: An interactive API Playground allows users to test queries and switch between lenses, regions, and file types.

Benchmarks

  • URL recall: 41% (Ranked #1 of 7 search APIs)

Use cases

  • AI agents use the API to find the correct source page with 41% recall across public benchmarks.
  • Developers extract clean Markdown content from URLs to provide clutter-free context for Large Language Models.
  • Research tools filter out AI-generated "slop" using SlopStop to ensure citations are based on human-created content.
  • Privacy-focused applications proxy search queries through Kagi to prevent upstream providers from tracking user identities.

Specifications

Published accuracy
35% fewer tokens per correct answer than the most accurate competitor tested
License
Licensed for commercial and AI use
Access
API, hosted MCP server, official SDKs (Go, Python, TypeScript, Rust)
Fine-tuning
Personalization via domain ranking, custom URL rules, and Lenses
Upstream Sources
12 providers per query
URL Recall Rate
41% (Benchmark leader)
Token Efficiency
35% fewer tokens per correct answer
Extract Limit
Up to 10 URLs per request
Response Formats
JSON, Markdown
Invoicing Threshold
Every 30 days or at $100 usage
SDK Availability
Go, Python, TypeScript, Rust

How it works

  1. Get API Keys: Developers sign up and generate API keys through the Kagi dashboard to begin integration.
  2. Configure Queries: Configure search queries with optional filters like Lenses, regions, dates, and file types in the API Playground or via SDKs.
  3. Parallel Processing and Ranking: The API fans out the query to 12 providers, merges results, and applies re-ranking and SlopStop filtering.
  4. Receive Clean Data: Receive structured JSON results or clean Markdown content ready for consumption by LLMs and AI agents.

Pricing

Search Requests
$12 per 1,000 searches
Extract Pages
$4 per 1,000 extracts

FAQ

Is the API private?

Every query is proxied through Kagi, ensuring that upstream providers never see who is asking the question, protecting user and agent privacy.

Can I use the API for commercial AI applications?

Yes, the Kagi Search API is specifically licensed for commercial and AI use, including research and compliance applications.

What is SlopStop?

SlopStop is a quality control feature powered by community reporting and auto-classification that flags and downranks AI-generated content to prioritize human work.

Get API keys

Added by AntSpace from public sources (kagi.com).

Related

More AI models by Kagi