AlphaAvatar DeepResearch Plugin

A modular deep-research and content acquisition middleware for AlphaAvatar, providing unified access to web search, iterative research, and content extraction.

This plugin enables AlphaAvatar agents to perform broad information gathering, exploratory research, and multi-step investigation workflows, without coupling agent logic to any specific search engine, crawler, or research provider.

DeepResearch focuses on acquiring and structuring external knowledge. Persistent storage and long-term reuse are handled by downstream tools such as the RAG plugin.

Features

• Unified Research Interface Abstracts search, browsing, and content extraction behind a single, agent-friendly API.

• Broad & Exploratory Research Designed for unfamiliar, complex, or open-ended topics that require multiple sources.

• Multi-step Research Support Supports iterative workflows such as search → read → refine → compare → synthesize.

• Source-aware Outputs Returns structured results including titles, URLs, snippets, and extracted content.

• Downstream-ready Artifacts Produces Markdown text and PDF files suitable for summarization, analysis, or indexing by RAG.

When to Use DeepResearch

Use this plugin when the task involves one or more of the following:

• Broad information gathering from multiple web sources
• Exploratory research on unfamiliar or complex topics
• Collecting background knowledge, trends, or comparisons
• Answering open-ended questions that cannot be resolved from a single source
• Acquiring external content that may later be indexed by a RAG system

Functionality

It exposes four core operations (op) that can be composed into a research pipeline:

• search Perform a lightweight web search for quick discovery. Use this when fast, broad results are needed with minimal reasoning.

• research Perform deep, multi-step research. Use this when the question requires decomposition, iterative searching, cross-source comparison, and reasoning.

• scrape Given a list of URLs, fetch and extract the main page contents, then merge them into a single Markdown text suitable for downstream processing (e.g., summarization or indexing).

• download Given a list of URLs, fetch pages and convert them into stored PDF artifacts, returning a list of file references for downstream tools or plugins (e.g., building a local RAG index).

Typical Workflow

1. Discover information using search or research

2. Acquire content using scrape or download

3. Pass outputs downstream

   • Summarize or analyze directly
   • Or hand off to the RAG plugin for indexing and long-term retrieval

Installation

pip install alpha-avatar-plugins-deepresearch

Supported DeepResearch Frameworks

Default: Tavily

Official Website

Tavily is a search and research API designed for LLM and agent workflows, emphasizing relevance, freshness, and machine-readable outputs, making it well-suited for autonomous research agents.

Additional DeepResearch backends can be integrated in the future without changing agent logic.