brave-search（勇敢的寻找）

Brave Search API

Overview

Search the web, images, and news using Brave's privacy-focused Search API. Also supports AI Grounding for cited answers via OpenAI SDK compatibility.

Prerequisites

API Key required. Get one at https://api-dashboard.search.brave.com

export BRAVE_API_KEY="your-api-key"

Free tier: 2,000 queries/month. Pro plans unlock Local POI search and AI Grounding.

When to Use

• Searching the web for current information
• Finding images on a topic
• Getting recent news articles
• AI-grounded answers with citations
• Autocomplete suggestions
• Location/POI searches (Pro plan)

High-Throughput Usage

Go wide, go fast. This API supports high concurrency—up to 50 requests/second. Don't hold back:

• Fire searches in parallel. Need to research 10 different topics? Launch 10 searches simultaneously. Use multiple Bash tool calls in a single message.
• Use subagents for heavy results. When expecting lots of context (many results, extra snippets, research mode), dispatch a subagent to run the search and synthesize findings. This preserves your main context window.
• Batch related queries. Searching for competitors, alternatives, or multiple aspects of a topic? Run them all at once.

digraph parallel_search {
    rankdir=LR;
    node [shape=box];

    task [label="Research task"];
    q1 [label="Query 1"];
    q2 [label="Query 2"];
    q3 [label="Query 3"];
    subagent [label="Subagent\n(preserves context)"];
    synthesize [label="Synthesized\nfindings"];

    task -> q1;
    task -> q2;
    task -> q3;
    q1 -> subagent [style=dashed];
    q2 -> subagent [style=dashed];
    q3 -> subagent [style=dashed];
    subagent -> synthesize;
}

When to use subagents:

• Searching for many sources on a topic (launch search subagent, return summary)
• Deep research with AI Grounding (high token usage, let subagent handle)
• Comparing multiple options (run parallel searches, subagent synthesizes)

When to search directly:

• Quick single queries where you need the raw URLs/snippets
• Low result counts where context isn't a concern

Quick Reference

API Endpoints

Common Parameters

Web Search

# Basic search
brave-search web "python async tutorial" --count 10

# Filter by freshness (pd=24h, pw=7d, pm=31d, py=365d)
brave-search web "latest news" --freshness pd

# Filter by country and language
brave-search web "local restaurants" --country US --lang en

# Safe search (off, moderate, strict)
brave-search web "query" --safesearch strict

# Get extra snippets
brave-search web "query" --extra-snippets

# Filter result types (web, news, videos, images, discussions)
brave-search web "query" --filter web,news

Image Search

# Basic image search
brave-search images "mountain sunset"

# With safe search
brave-search images "landscape" --safesearch strict --count 20

News Search

# Recent news
brave-search news "AI developments" --count 10

# News with freshness filter
brave-search news "election results" --freshness pd

AI Grounding (Cited Answers)

# Get an AI answer with citations
brave-search ai "What is the tallest building in the world?"

# Enable deep research (multiple searches, slower)
brave-search ai "Compare React and Vue in 2024" --research

Workflow

digraph brave_search {
    rankdir=TB;
    node [shape=box];

    need [label="What do you need?" shape=diamond];
    web [label="Web Search\nbrave.py web"];
    images [label="Image Search\nbrave.py images"];
    news [label="News Search\nbrave.py news"];
    ai [label="AI Answer\nbrave.py ai"];

    results [label="Parse JSON results"];
    cite [label="Include citations\nfor AI answers"];

    need -> web [label="web pages"];
    need -> images [label="images"];
    need -> news [label="recent news"];
    need -> ai [label="cited answer"];

    web -> results;
    images -> results;
    news -> results;
    ai -> cite;
}

Response Structure

Web Search Results

{
  "web": {
    "results": [
      {
        "title": "Page Title",
        "url": "https://example.com",
        "description": "Snippet from the page...",
        "extra_snippets": ["Additional context..."]
      }
    ]
  },
  "query": {
    "original": "search query",
    "altered": "modified query if spellchecked"
  }
}

AI Grounding Response

Returns OpenAI-compatible format with inline citations:

The tallest building is the Burj Khalifa[1] at 828 meters...

[1] https://source-url.com

Common Options

Error Handling

Rate Limits

• Free: 1 req/sec, 2,000/month
• Pro: Higher limits, check dashboard
• Response headers show remaining quota: X-RateLimit-Remaining

Common Mistakes