← Back to Dashboard
1. Function and API Tool Patterns2. Retries, Idempotency, and Timeouts3. Web Search: Built-In Server-Side Grounding
Web Search: Built-In Server-Side Grounding
📚 Tool Use and Orchestration⏱ 9 min⭐ 85 XP
Grounding Without a Third-Party Search Vendor
Amazon Bedrock now offers Web Search (general availability, August 2026) - a built-in, server-side tool that lets a model ground its answers in current web knowledge entirely inside AWS, with zero data egress to an external search provider. Previously, adding web grounding meant onboarding a third-party search API, managing separate keys and billing, building custom retrieval orchestration, and running a security review for every vendor. Web Search removes all of that: it is a single parameter on an existing API call.
How It Works
- Amazon operates the web index and knowledge graph - tens of billions of documents, refreshed continually, informed by Amazon's experience running Alexa+, Amazon Quick, and Kiro.
- Bedrock performs semantic snippet extraction server-side rather than returning raw pages, keeping results context-efficient for the model's context window.
- It integrates through the same standard tool-use interface you already use for custom tools, and is compatible with the OpenAI Responses API on Bedrock.
- A single API call returns a grounded response with citations - Bedrock manages the entire search lifecycle server-side.
Web Search vs. a Custom Retrieval Tool
| Aspect | Custom search tool (self-integrated) | Bedrock Web Search |
|---|---|---|
| Vendor onboarding | Required per provider | None - built into Bedrock |
| Data egress | Requests leave AWS to the vendor | Zero egress - stays inside your AWS environment |
| Security review | New review per external API | Covered by your existing Bedrock/AWS review |
| Integration effort | Custom orchestration code | One parameter on an existing API call |
At GA, Web Search is available in US East (N. Virginia), US East (Ohio), and US West (Oregon), for OpenAI models (GPT-5.4, GPT-5.5, and the GPT-5.6 family) via the Responses API.
Key idea: Web Search is a grounding tool, not a knowledge base replacement - use it for current-events/freshness gaps, and keep Knowledge Bases + RAG for your own proprietary or contractual documents.
🧪 Knowledge Check
Press 1-4 to select1 of 3
What is the main advantage of Bedrock's built-in Web Search tool over a self-integrated third-party search API?
It is always more accurate
No vendor onboarding, no external API keys, and zero data egress from AWS
It replaces Knowledge Bases entirely
It requires no IAM permissions at all