AWS AgentCore Web Search API GA — Two Approaches to Agent Web Search
AWS has made the Web Search API on Amazon Bedrock AgentCore generally available. Introduced in the AWS News Blog weekly roundup (November 9, 2026) by Daniel Abib, it complements the Web Answer tool as a second approach to agent-driven web search capabilities.
Web Search vs Web Answer: Different Design Philosophies
AgentCore offers two web search-related tools with distinct design philosophies:
- Web Answer: Returns a synthesized, cited answer to a natural-language question in a single tool call. It plans its own searches, runs several in parallel, and decides when it has enough evidence. The agent receives one consolidated answer
- Web Search: Returns ranked search results for the agent's model to synthesize directly. Agents receive raw results and have full control over how to process, analyze, and combine them
In other words, Web Answer is a black box that handles "search + synthesis" in one step, while Web Search provides just the search results — like building blocks. Depending on the agent architecture, teams can choose either or both.
Technical Features
Key features of the Web Search API:
- Attached as a target on AgentCore Gateway
- Called via MCP (Model Context Protocol)
- Domain include/deny lists to restrict search scope
- Publish-date filters for freshness control
- Available in US East (N. Virginia)
Web Search alongside AgentCore Runtime V2
The same weekly roundup also covered the new AgentCore Runtime V2, which introduces on-demand session memory paging and snapshot-based cold start optimization. P75 cold starts dropped to approximately 2 seconds for 200MB images, compared to ~5.4 seconds on the original runtime.
Web Search operates on this optimized Runtime V2 infrastructure, ensuring the full pipeline — from search retrieval through agent-side synthesis — runs efficiently.
Conclusion
AWS AgentCore's Web Search API GA adds flexibility to agent architecture design. Where Web Answer automates "search through synthesis," Web Search opens up the raw results so the agent's model can process them however it sees fit. Used together, they enable sophisticated search strategies across complex workflows.
As agents evolve from single-step question-answering to multi-step information gathering and synthesis, AWS is expanding the AgentCore ecosystem to support both approaches.