Research

The research tool performs multi-step web research by chaining web search, URL extraction, content fetching, and LLM summarization.

How It Works

The research pipeline:

1. Web Search — Searches for the query using the configured search provider
2. URL Extraction — Extracts the most relevant URLs from search results
3. Content Fetching — Downloads and parses the content from each URL
4. LLM Summarization — Summarizes the gathered information into a coherent response

This produces much more thorough results than a single web search, as it actually reads and synthesizes the content of multiple pages.

Usage

User: Research the current state of quantum computing in 2024
Agent: [uses research tool]
  → web_search: "quantum computing 2024 progress"
  → fetch: https://example.com/quantum-2024
  → fetch: https://another.com/quantum-advances
  → summarize: "Based on multiple sources..."

Configuration

The research tool uses whatever search provider is configured:

export POCKETCLAW_WEB_SEARCH_PROVIDER="tavily"
export POCKETCLAW_TAVILY_API_KEY="tvly-..."

Tool Schema

{
  "name": "research",
  "description": "Perform deep research on a topic by searching, reading, and synthesizing multiple sources",
  "input_schema": {
    "type": "object",
    "properties": {
      "query": {
        "type": "string",
        "description": "The research topic or question"
      },
      "depth": {
        "type": "integer",
        "description": "Number of sources to read (default: 3)"
      }
    },
    "required": ["query"]
  }
}

Policy Group

Belongs to group:research.