khojkar

khojkar (pa: ਖੋਜਕਾਰ, ipa: /kʰoːd͡ʒ.kɑːɾ/) is a deep research agent.

Installation

Prerequisites

• Python 3.12
• uv - Fast Python package installer and resolver

Quick Start (PyPI)

The easiest way to run khojkar is directly using uvx, which will fetch it from PyPI:

# Run directly from PyPI using uvx
uvx khojkar research --topic "Your research topic" --output report.md

To set up credentials when using this method, create a .env file in your current working directory with your API keys. uvx will automatically load environment variables from a .env file.

Setup Credentials

khojkar uses LLMs via the LiteLLM library, which requires API keys for the models you want to use.

1. Create a .env file in your current working directory:

   touch .env

2. Add your API keys to the .env file. You only need the key for the model you intend to use. The default is gemini/gemini-2.0-flash, which requires the GEMINI_API_KEY. For other providers like OpenAI or Anthropic, you would use OPENAI_API_KEY or ANTHROPIC_API_KEY respectively. You also need to add credentials for the Google Programmable Search Engine: SEARCH_ENGINE_ID and SEARCH_ENGINE_API_KEY.

   # Required for the default model (gemini/gemini-2.0-flash)
   GEMINI_API_KEY=your_gemini_api_key
   
   # Required for Google Programmable Search Engine
   SEARCH_ENGINE_ID=your_search_engine_id
   SEARCH_ENGINE_API_KEY=your_search_engine_api_key
   
   # Optional: add other LLM API keys as needed
   # OPENAI_API_KEY=your_openai_api_key
   # ANTHROPIC_API_KEY=your_anthropic_api_key

   Note: Refer to the LiteLLM documentation for the specific environment variable names required for different LLM providers.

Usage

Basic Usage:

khojkar research --topic "Your research topic" --output report.md

Advanced Options:

# Use a different model
khojkar research --topic "Your research topic" --model "openai/gpt-4o" --output report.md

# Limit research steps
khojkar research --topic "Your research topic" --max-steps 5 --output report.md

# Use multi-agent research mode
khojkar research --topic "Your research topic" --multi-agent --output report.md

License

See LICENSE file for details.