AWS announced general availability of Web Answer on Amazon Bedrock AgentCore on November 5, 2026. It is a managed tool for an AI agent that needs an answer grounded in current web pages. Instead of receiving search results and building its own retrieval and synthesis chain, the agent asks a natural-language question through one tool call and receives answer text with source citations. AWS describes the workflow; VIBPS has not run it or evaluated answer quality.
How does AgentCore Web Answer work?
A developer adds the AWS-owned Web Answer connector as a target on an existing AgentCore Gateway. The Gateway authenticates the agent call and exposes the target as a Model Context Protocol tool. Web Answer searches an Amazon-operated web index, reads selected pages and writes an answer whose claims map to source URLs. The tool returns readable markdown as well as structured citation data. A cited result is easier to inspect, but a citation alone does not prove that an answer interpreted the source correctly.
AWS distinguishes this from its Web Search tool. Web Search returns ranked links and metadata for the agent's own model to assess; Web Answer performs retrieval and synthesis inside the managed tool. A maker choosing between them is choosing where source selection and final wording happen. Web Answer can suit a factual subquestion in a larger app, while raw search results offer more control over how the app weighs evidence.
What must a developer set up, and where is it available?
The release walkthrough starts with an AWS account, an AgentCore Gateway in US East (N. Virginia), appropriate IAM permissions and a Gateway service role allowed to invoke the connector. The agent then discovers the target through tools/list and calls it through tools/call. The current AWS documentation lists us-east-1 as the connector's available region. It also describes target and per-request domain controls, date filters and an optional country; the filters depend on the connector version and do not let an agent widen an administrator's domain restrictions.
AWS says a question stays within AWS infrastructure and is not sent to a third-party search engine. This describes its processing path, not a guarantee that every web source is correct or that an agent should act on an answer without review. The developer still needs to decide which questions may reach the tool, how citations are shown to users, and what happens when the evidence is weak.
What does Web Answer cost?
AWS's November 5 post lists $10 per 1,000 Web Answer calls, or $0.01 per call at that quoted rate. A Gateway and other AWS resources can add charges; the walkthrough explicitly warns that its setup creates billable resources. A maker should check current AWS pricing and model the number of questions, retries and adjacent agent calls before treating the quoted tool rate as the cost of a whole feature.
For a software creator, the practical release is a new way to put current, source-linked answers inside an app without maintaining a separate search and citation pipeline. The hard product work remains: showing sources clearly, recognizing uncertainty and keeping the user's trust when an answer is incomplete. That distinction is relevant to VIBPS's maker audience, whose apps increasingly combine existing services into new experiences.