Microsoft Web IQ Search
Use search_provider="webiq" with a Microsoft Web IQ API key. This integration calls the Web Search v3 REST API, which is separate from Grounding with Bing
Set WEBIQ_API_KEY in the process environment or pass api_key explicitly. The provider sends it in the x-apikey header. This initial integration uses API-key authentication; it does not acquire or refresh Entra ID tokens
Python SDK
import asyncio
from litellm import asearch, search
result = search(
query="retrieval augmented generation research",
search_provider="webiq",
max_results=5,
country="US",
)
print(result.model_dump())
async def main():
result = await asearch(query="retrieval augmented generation research", search_provider="webiq")
print(result.model_dump())
asyncio.run(main())
The provider requests contentFormat="passage" and maxLength=5000 by default. content becomes the standard result's snippet, while lastUpdatedAt supplies date and last_updated when available. Crawl timestamps are preserved as crawledAt, not presented as publication dates. Other response metadata, including traceId and instrumentation fields, is preserved unless its key conflicts with a standard LiteLLM response field; the integration does not send instrumentation pings
Gateway
Add this search tool to your existing proxy configuration:
search_tools:
- search_tool_name: webiq-search
litellm_params:
search_provider: webiq
api_key: os.environ/WEBIQ_API_KEY
Set LITELLM_PROXY_URL to your existing LiteLLM deployment URL, without a trailing slash, and LITELLM_API_KEY to your LiteLLM gateway key. The gateway key authenticates your application to LiteLLM; WEBIQ_API_KEY authenticates the gateway to https://api.microsoft.ai/v3/search/web
curl "$LITELLM_PROXY_URL/v1/search/webiq-search" \
-H "Authorization: Bearer $LITELLM_API_KEY" \
-H "Content-Type: application/json" \
-d '{"query":"retrieval augmented generation research","max_results":5}'
You can also send POST /v1/search or POST /search with search_tool_name in the request body:
{
"search_tool_name": "webiq-search",
"query": "retrieval augmented generation research",
"max_results": 5
}
Search calls follow the gateway's search tool access rules. If your deployment restricts search tools, grant the caller access to webiq-search; the same rules apply to web-search interception
Dashboard
With a database configured for the proxy, sign in as a proxy administrator and open Search Tools. Click Add New Search Tool, enter webiq-search as the Search Tool Name, select Microsoft Web IQ as the Search Provider, and enter your Web IQ key in API Key. Leave Cost per Query (USD, Optional) empty to use LiteLLM's catalog price, or set your negotiated rate
Click Test Connection, close the test results, and click Add Search Tool to save. Open the saved tool, enter a query, and click Search to inspect the returned titles, URLs, and passages. Use the saved tool's name in the gateway requests above; a dashboard-created tool does not also need an entry in config.yaml
Router
For SDK applications that manage named search tools, configure Web IQ on a Router and call router.asearch with the tool name:
import asyncio
import os
from litellm import Router
router = Router(
model_list=[],
search_tools=[
{
"search_tool_name": "webiq-search",
"litellm_params": {
"search_provider": "webiq",
"api_key": os.environ["WEBIQ_API_KEY"],
},
}
],
)
async def main():
result = await router.asearch(
search_tool_name="webiq-search",
query="retrieval augmented generation research",
max_results=5,
)
print(result.model_dump())
asyncio.run(main())
Web-search interception
To select Web IQ for web-search interception, merge these settings into the proxy configuration alongside the webiq-search tool. Set enabled_providers to the model providers used by your application; this example selects Bedrock
litellm_settings:
callbacks:
- websearch_interception
websearch_interception_params:
search_tool_name: webiq-search
enabled_providers: ["bedrock"]
Send a model request with a supported web-search tool, as shown in the interception usage examples. Enabling the callback alone does not add web search to every model request. LiteLLM routes intercepted searches to webiq-search and uses its existing tool conversion, streaming, and model continuation behavior
Parameters
| LiteLLM parameter | Web IQ request |
|---|---|
query | query; a list of queries is joined with spaces |
max_results | maxResults |
country | region, uppercased |
search_domain_filter | site: query operators, with -domain entries mapped to -site: exclusions |
max_tokens_per_page | Not forwarded; Web IQ's maxLength measures characters, not tokens |
Pass native options such as language, region, location, contentFormat, maxLength, maxResults, and safeSearch as keyword arguments. Explicit native values override mapped values and integration defaults. Web IQ validates its own parameter limits
Domain filters narrow search relevance. Microsoft also notes that site: queries can return adult content regardless of safeSearch; see the API reference before relying on domain filtering in a restricted-content application
The default API base is https://api.microsoft.ai/v3. Set api_base in the SDK call or proxy search configuration to select a different base, with /search/web appended once. HTTPS is required for every endpoint, including custom endpoints with an explicit key. HTTP endpoints are rejected before credentials are resolved or a request is sent. A caller-selected host cannot receive a server-managed API key unless the host matches the default; pass an explicit api_key with a custom endpoint
LiteLLM registers webiq/search at $0.007 per query. Microsoft's published pricing lists /web at $7 per 1,000 API calls, effective October 16, 2026. Live Web IQ validation requires an enabled account and a valid API key
This integration currently supports /web. The /video, /news, /images, and /browse endpoints are outside its current scope