Exa

Exa (formerly Metaphor)

Overview

Exa is a Neural Search Engine built specifically for AI agents and LLMs. Unlike traditional search engines that rely on keyword frequency (TF-IDF/BM25), Exa uses semantic embeddings to find content based on the meaning and context of a query.

Key Capabilities

• Semantic Mapping: Finds “more like this” by providing a URL as a starting point.
• Agent-Native Synthesis: Returns structured JSON using outputSchema, mapping claims directly to source URLs (grounding).
• High-Fidelity Scrapping: Integrated contents engine that converts HTML to clean, LLM-optimized Markdown.

Role in Digital Brain

In the Agentic Architecture, Exa powers the Explorer role. It is used to:

1. Find “Shadow Knowledge”: Discover obscure technical gists, personal blogs, and research that standard search misses.
2. Generate Related Reading: Automatically populate “Related Concepts” sections in the wiki during ingestion.
3. Verify Stale Claims: Use deep-reasoning search to check if a documented technical claim still reflects current industry standards.

Reference

• MCP URL: https://mcp.exa.ai/mcp
• Documentation: Exa-API-Setup-Guide