Skip to main content

Exa

Category: AI Infrastructure

A neural search engine built from the ground up for AI agents and LLMs, using embedding-based retrieval to understand semantic meaning rather than keyword matching. Exa was founded in 2021. The company is led by Will Bryk. Based in San Francisco, USA. Team size: 82. Total funding raised: $111.0M. Latest round: Series B. Key investors include ["Benchmark","NVentures (Nvidia)","Lightspeed Venture Partners","Y Combinator","Caffeinated Capital"].

Founded
2021
Headquarters
San Francisco, USA
Team size
82
Total funding
$111.0M

Value proposition

Provides high-precision, semantically relevant web data for AI models, reducing hallucinations and improving the quality of RAG (Retrieval-Augmented Generation) systems through clean, machine-ready content.

Products and solutions

["Exa API (Search, Contents, and Find Similar endpoints)","Exa Instant (Sub-200ms low-latency neural search for real-time agents)","Websets (Large-scale data enrichment and targeted web-scale query handling)","Exa Web Search (Semantic search UI for human and AI hybrid research)"]

Unique value

Unlike traditional engines that use keyword indexing and PageRank, Exa uses a 'link prediction transformer' architecture that treats the web like a giant language model to predict relevant content.

Target customer

AI developers, LLM labs (e.g., OpenAI, Anthropic), enterprise AI teams, and startups building agentic workflows.

Industries served

["Artificial Intelligence & Machine Learning","Software Development","Data Science & Analytics","Market Research","Enterprise Automation","Cybersecurity"]

Technology advantage

Proprietary end-to-end neural stack that indexes billions of web documents as embeddings, allowing for 'Find Similar' capabilities and complex natural language queries that keyword-based engines cannot resolve.

How they differentiate

Exa uses a proprietary 'link prediction transformer' architecture that indexes the web as embeddings rather than keywords, allowing it to understand semantic intent and return clean, LLM-ready content (Markdown/HTML) optimized for RAG systems.

Main competitors

["Perplexity AI","Tavily","Google Vertex AI Search"]

Key partnerships

["Nvidia (NVentures) - Strategic investor and infrastructure partner","Fireworks AI - Collaboration on high-performance AI inference and search workflows","Y Combinator - Alumnus (W21)","Digital Realty - Partnership for global connectivity and data center infrastructure"]

Notable customers

["OpenAI","Anthropic","Cresta","Fireworks AI","Databricks"]

Major milestones

["Rebranded from Metaphor Systems to Exa in 2023","Raised $17M Series A led by Lightspeed in July 2024","Secured $85M Series B led by Benchmark at a $700M valuation in September 2025","Launched Exa Instant, achieving sub-200ms latency for real-time AI agents"]

Growth metrics

Reported $12M ARR in 2025; team size of approximately 82 employees.

Market positioning

Infrastructure-level neural search engine for AI developers and agentic workflows.

Geographic focus

Global (API-based), with primary operations in North America.

Patents and IP

Proprietary neural indexing and representation learning techniques; no specific public patent registrations disclosed.

About Will Bryk

Will Bryk was an early engineer at Cresta, an AI startup focused on real-time intelligence for contact centers. He studied Physics and Computer Science at Harvard University, where he led the robotics club and conducted research on human-AI interaction. He also held internships at SpaceX and Zoox.

Official website: