Skip to main content

Web Search

Some models available through BluesMinds support live web search, allowing the model to retrieve up-to-date information before generating a response.

Web search is available on select models that have built-in search capability or via tool-augmented search. Check the current model list:

curl https://api.bluesminds.com/v1/models \
  -H "Authorization: Bearer $BLUESMINDS_API_KEY"

Models known to support search include:

  • grok-3 (xAI — native web search)
  • gemini-2.0-flash (Google — grounding with Google Search)
  • Select search-augmented variants

Using Search via Tool Calling​

For models that don't have built-in search, you can implement web search as a tool:

from openai import OpenAI
import json, requests

client = OpenAI(
    api_key="sk-your-api-key",
    base_url="https://api.bluesminds.com/v1",
)

tools = [{
    "type": "function",
    "function": {
        "name": "web_search",
        "description": "Search the web for current information",
        "parameters": {
            "type": "object",
            "properties": {
                "query": {
                    "type": "string",
                    "description": "The search query"
                }
            },
            "required": ["query"]
        }
    }
}]

messages = [{"role": "user", "content": "What is the current price of Bitcoin?"}]

response = client.chat.completions.create(
    model="gpt-4o",
    messages=messages,
    tools=tools,
)

msg = response.choices[0].message

if msg.tool_calls:
    tool_call = msg.tool_calls[0]
    query = json.loads(tool_call.function.arguments)["query"]

    # Execute your search (e.g., using SerpAPI, Brave Search, etc.)
    search_result = your_search_function(query)

    messages.append(msg)
    messages.append({
        "role": "tool",
        "tool_call_id": tool_call.id,
        "content": json.dumps(search_result),
    })

    final = client.chat.completions.create(model="gpt-4o", messages=messages)
    print(final.choices[0].message.content)

grok-3 has built-in web search that activates automatically for time-sensitive queries:

{
  "model": "grok-3",
  "messages": [
    {"role": "user", "content": "What happened in the tech industry this week?"}
  ]
}

Grok will search the web and include sources in its response.

Using Gemini with Google Grounding​

When using Gemini models via the /v1beta/models path, Google Search grounding can be enabled. For the standard /v1/chat/completions endpoint, use Gemini's standard capabilities and supplement with tool-based search.

Best Practices​

  • Use search for time-sensitive queries — LLMs have training cutoffs; search retrieves current info
  • Cite sources — ask the model to include URLs from search results for verifiability
  • Rate limit awareness — search tool calls count against your RPM limit
  • Cache common queries — if many users ask the same question, cache search results to reduce latency and cost

Limitations​

  • Native search is only available on specific models (Grok, Gemini variants)
  • Tool-based search requires you to integrate a search API (SerpAPI, Brave, Bing, etc.)
  • Search results are injected into the context window — very long results may exceed context limits