Overview

Auto-decomposes one query into many parallel subqueries. Ready for production.

62ms P50 latency · Scales to 1M+ QPS · Broader coverage in a single call.

Works best for research agents, hard questions, and comprehensive lookup.

Ranked web results, highlights, and summaries. Ready for generation.

62ms P50 latency · Scales to 1M+ QPS

Works best with structured queries generated by LLMs.

Image retrieval across the open web. Ready for design and creative workflows.

Works best for design references, asset lookup, and content creation.

Video retrieval across the open web. Ready for media discovery and research.

Works best for video discovery, tutorial lookup, and clip sourcing.

One API to call any LLM, with unified billing. Ready for production.

OpenAI-compatible · Switch providers without rewriting code.

Works best for multi-model apps, fallback chains, and cost routing.

Clean markdown, intent-focused highlights, and page classification. Ready for ingestion.

Clean markdown extraction · success-only pricing

Works best for AI agents, RAG pipelines, and content monitoring.

Vector embeddings for retrieval, similarity, and ranking. Ready for production.

#1 on RTEB (0.7953) · Strong multilingual performance.

Works best for RAG, semantic search, and recommendations.

Multimodal embeddings for cross-modal retrieval, similarity, and ranking.

#1 on MMEB-V2 (80.09) · Unified vision-language space.

Works best for multimodal RAG and image-text retrieval.