Metaphor
FreeLanguage model powered search.
FreeFree tier
About Metaphor
Exa (formerly Metaphor) is an applied AI research lab building a high-performance search API for AI agents. It provides a unified API for web search, crawling, deep research, and content monitoring, optimized for token efficiency and low latency. Exa powers coding agents, chatbots, news monitoring, and more with structured outputs and grounded citations. It indexes billions of web pages and uses state-of-the-art neural retrieval models. The API offers configurable latency from 180ms to 1s for fast searches or deeper multi-step research.
Key Features
Unified API for search, crawling, and deep research
Token-efficient highlights for LLM context
Configurable latency from 180ms to 1s
Structured outputs (JSON) with citations
Monitors for tracking new events across the web
Deep search with multi-step agent workflows
Low-latency search for coding agents (docs, repos, Stack Overflow)
SOTA neural embedding models trained on exa-cluster
Index of 500B+ webpages with real-time updates
Pros & Cons
Pros
- Best-in-class accuracy on FRAMES, Tip-of-Tongue, and Seal0 benchmarks
- Token-efficient highlights reduce LLM costs (up to 90% token reduction)
- Flexible latency options for speed vs. depth trade-offs
- Supports structured output extraction from web content
- Free tier with $20 credits on sign-up and $10 monthly credits
- Used by major companies like Cursor, Cognition, HubSpot
Cons
- Pricing can be complex with multiple endpoint tiers and additional costs for extra results
- Not open-source; proprietary API service
- Learning curve for configuring different search modes and effort levels
Best For
Coding agents searching docs, repos, and changelogsChatbots grounding responses with web contentNews monitoring and event trackingCompany and people enrichment (structured data extraction)Voice agents needing real-time web dataDeep research with structured outputs and citationsBuilding AI-powered search applications
FAQ
What is Exa?
Exa is an applied AI research lab building a modern search engine API for AI agents. It provides search, crawling, deep research, and monitoring capabilities optimized for token efficiency and latency.
How does pricing work?
Exa offers a free tier with $20 credits on sign-up and $10 in monthly credits. Paid plans start at $7 per 1,000 requests for search, with additional costs for deep search, monitors, and contents. Agent pricing is based on compute units and fixed effort modes.
Is Exa open source?
No, Exa is a proprietary API service. The website does not indicate open-source availability.
What latency can I expect?
Search latency is configurable, ranging from under 200ms for instant responses up to 1 second for deeper retrieval. Deep research (Agent mode) can take longer depending on effort level.
What types of content can Exa retrieve?
Exa can retrieve full webpage contents, token-efficient highlights, and structured data (JSON) from web pages. It also supports content from GitHub repos, docs, and other developer resources.