Foundry Web Search — Bing Grounding (Responses API)

You have access to real-time web search via the Foundry Responses API bing_grounding tool. Results come with inline URL citations. No sandbox egress policy exceptions needed — search runs server-side.

Note: Requires a Bing Grounding connection configured in the Foundry project. If not available, the model will answer from its training data instead.

Endpoint

All requests: http://localhost:8443 with ?api-version=2025-11-15-preview. Auth is automatic.

Operations

Search the web

curl -s -X POST 'http://localhost:8443/openai/responses?api-version=2025-11-15-preview' \
  -H 'Content-Type: application/json' \
  -d '{"model":"gpt-4.1","input":"What are the latest Azure AI Foundry announcements?","tools":[{"type":"bing_grounding","bing_grounding":{"search_configurations":[{"project_connection_id":"AUTO_DISCOVERED_AT_RUNTIME"}]}}],"store":false}'

The response includes the answer with inline URL citations.

Simple search (no Bing connection needed)

If Bing is not configured, use plain chat — the model answers from training data:

curl -s -X POST 'http://localhost:8443/v1/chat/completions' \
  -H 'Content-Type: application/json' \
  -d '{"model":"gpt-4.1","messages":[{"role":"user","content":"What are the main features of Azure AI Foundry?"}]}'

When to use

• User asks about current events, news, recent changes
• Need real-time data: weather, stock prices, sports scores
• Verifying facts with web sources
• User says "search the web", "look up", "what's the latest"

When NOT to use

• For information in uploaded documents (use foundry-knowledge)
• For calculations or data analysis (use foundry-code)
• For recalling previous conversations (use foundry-memory)