TechPrufer
Sign inGet started

Blogs›Exa›Exa Agent

Exa Agent

Exa Agent

8
posts
2025–2026

Exa has launched exa-code, a web-scale context tool for coding agents. It leverages Exa's search engine and prioritizes code examples to provide highly relevant and dense context, typically a few hundred tokens. The system hybrid searches over 1B+ webpages, extracts and reranks code examples using an ensemble method, and returns either concise code examples or full documentation pages. Evaluations show exa-code significantly reduces LLM hallucinations in coding tasks by providing optimized code example recall from a dedicated index sourced from GitHub and Exa's web index, using code-specific retrieval models. The exa-code MCP is available via Smithery and Exa's documentation.

2026

Introducing Exa Agent

6/16/2026

Introduced Exa Agent, a new API that integrates top language models with Exa's web search tools for advanced web research. Detailed its capabilities in deep research, list-building, and entity enrichment, highlighting its ability to divide tasks into subtasks and use model fusion for cost-effectiveness. Quantified performance using benchmarks and introduced the WideSearch methodology for evaluating agent performance. Described how teams are using Exa Agent for finance and go-to-market tasks, providing a concrete example of company research. Outlined API effort levels and cost structures, and mentioned structured outputs and custom data input.

Introducing Deep Max: State-of-the-Art Agentic Search

4/20/2026

Introduced Deep Max, a new agentic search endpoint that leverages frontier LLMs and parallel calls to Exa Search. Detailed the technical reasons for its speed, including parallel tool calls, token-efficient content processing, and the performance of Exa's internal search stack. Benchmarked Deep Max against competitors on accuracy and latency across multiple agentic search evals.

WebCode: Search Evals for Coding Agents

3/23/2026

This post introduces WebCode, a new set of coding evaluations for web search for coding agents. It details methods for evaluating extraction faithfulness against golden references, using both LLM-judged and deterministic metrics. It also introduces a discriminative evaluation framework for RAG to assess groundedness independently of synthesis, demonstrating its importance in isolating search provider capabilities. The post details the generation of question-answer pairs from long-context documentation and the evaluation of retrieval quality on the entire web, measuring groundedness and citation precision.

Introducing Exa Deep: An Agent for Every Search

3/4/2026

Introduces Exa Deep, a significantly improved agentic search endpoint. Details its architecture which combines Exa Instant search with LLM reasoning for parallel agent generation and result synthesis. Highlights performance improvements (faster, cheaper) and new features like structured outputs with field-level grounding. Presents benchmark results comparing Exa Deep against competitors on HLE-Search, Deep Search QA, and FRAMES. Outlines key use cases including financial research, scientific literature, and news monitoring. Provides API usage details for `deep` and `deep-reasoning` types and pricing information.

2025

Introducing Exa 2.1

11/24/2025

This post announces Exa 2.1, detailing significant improvements to Exa's search APIs. It highlights the scaling of compute resources, leading to sub-500ms latency for Exa Fast and enhanced accuracy for Exa Deep through agentic search. New evaluation methodologies for both fast and agentic search are described, focusing on LLM grading of search results. The post also reiterates Exa's commitment to building an independent search engine from scratch, including semantic+lexical databases and large-scale crawling infrastructure.

Introducing Exa 2.0

10/10/2025

Introduced Exa 2.0 with three major updates: Exa Fast (sub-350ms P50 latency), Exa Auto (improved quality), and Exa Deep (agentic, high-quality search). Detailed the technical underpinnings including a larger index, new embedding model training on a 144x H200 cluster, and vector database optimizations (clustering algorithms, lexical compression, assembly optimizations in Rust). Outlined latency and search quality evaluation methods, including new benchmarks and RAG evaluation harnesses.

Use exa-code: Fast, efficient web context for coding agents

9/25/2025

Introduced exa-code, a new tool for coding agents that provides fast, efficient web context. The system works by hybrid searching webpages, extracting and reranking code examples for relevance, and returning concise context. Evaluations demonstrate exa-code's effectiveness in reducing LLM hallucinations for coding tasks by optimizing for code example recall.

Exa Raises $85M to Build the Search Engine for AIs

9/3/2025

This post announces Series B funding and outlines Exa's strategic direction as the 'search engine for AIs'. It details the technical differentiators of Exa's search for AI applications: high-quality knowledge (no ads, optimized ranking), full content access, low latency (sub-450ms API), high-compute options (Websets), customization capabilities, and Zero Data Retention (ZDR). It also highlights plans to scale indexing/processing, increase GPU cluster size for research, and expand the team to achieve 'perfect search'.