AWS released an Exa search integration for its Strands Agents SDK, enabling developers to build AI agents that autonomously execute web searches and retrieve information. The integration, available through the strands-agents-tools package, provides two primary capabilities: exa_search for semantic querying across categories like news and research papers, and exa_get_contents for direct, structured text extraction from specific URLs.

The search tool includes four modes to balance speed and quality: Instant (~200ms) for real-time needs, Fast (~450ms), Auto (~1s), and Deep (3-6s) for comprehensive research. Because Exa returns clean content formatted specifically for LLM context windows, agents avoid the processing overhead of cleaning raw HTML or ad-heavy search snippets.

This architecture allows for complex, multi-step agent workflows. For example, a deep research agent can execute a 6-step pipeline that automatically performs targeted searches, filters by date or source type, triggers live crawling for fresh content, and synthesizes data into a structured research brief. The underlying model dynamically decides when to invoke these tools and how to manage the reasoning loop based on the task.

AWS released an Exa search integration for its Strands Agents SDK, enabling developers to build AI agents that autonomously execute web searches and retrieve information. The integration, available through the strands-agents-tools package, provides two primary capabilities: exa_search for semantic querying across categories like news and research papers, and exa_get_contents for direct, structured text extraction from specific URLs.

The search tool includes four modes to balance speed and quality: Instant (~200ms) for real-time needs, Fast (~450ms), Auto (~1s), and Deep (3-6s) for comprehensive research. Because Exa returns clean content formatted specifically for LLM context windows, agents avoid the processing overhead of cleaning raw HTML or ad-heavy search snippets.

This architecture allows for complex, multi-step agent workflows. For example, a deep research agent can execute a 6-step pipeline that automatically performs targeted searches, filters by date or source type, triggers live crawling for fresh content, and synthesizes data into a structured research brief. The underlying model dynamically decides when to invoke these tools and how to manage the reasoning loop based on the task.