Web Search

Overview

The web search feature allows the AI agent to search the internet for relevant information while working on your tasks. This is particularly useful for:

• Finding documentation for libraries and APIs
• Looking up error messages and solutions
• Discovering best practices and examples
• Getting current information about packages

How It Works

Web search is implemented using OpenAI's built-in webSearchTool from the @openai/agents framework. The implementation is in apps/thesis/src/server/tools/web-search.tool.ts.

When the agent decides to search:

1. It formulates a search query based on context
2. OpenAI's search API retrieves relevant results
3. The agent processes and summarizes the findings
4. Results are incorporated into the response

Use Cases

Finding Documentation

thesis exec "How do I use pandas groupby with multiple aggregations?" --mode ask

The agent will search for pandas documentation and provide examples from official sources.

Debugging Errors

thesis exec "I'm getting 'CUDA out of memory' - what are the solutions?" --mode ask

The agent searches for common solutions and best practices for memory management.

Library Discovery

thesis exec "What's the best library for plotting time series in Python?" --mode ask

The agent researches current options and provides recommendations with pros/cons.

Configuration

Web search is available in all three agent modes (Ask, Plan, Agent) but can be disabled if needed.

The tool availability is controlled by the webSearchEnabled flag in apps/thesis/src/server/libs/agent.tools.ts.