Deep research

Aesoperator's deep research capability performs comprehensive research by combining browser automation, data extraction, and memory-powered analysis. Research tasks typically take 2-8 hours depending on depth and scope.

How It Works

The research process:

1. Takes a research topic/question and constraints as input

2. Uses Firefox to navigate and scrape relevant sources

3. Extracts and processes information using vision and language models

4. Builds a knowledge graph in pgvector for semantic search

5. Generates insights using LLM analysis

For example, researching "Latest advances in fusion energy":

1. Crawls scientific papers, news articles, and research lab websites

   • Uses Firefox with Selenium for web navigation

   • Accesses arXiv, Google Scholar, ScienceDirect via APIs

   • Downloads PDFs and HTML content for processing

2. Extracts key findings about recent breakthroughs and technical progress

   • Uses newsonnet Vision to analyze diagrams and figures

   • Uses newsonnet Vision to do any computer vision tasks

3. Builds knowledge graph connecting research teams, technologies, and results

   • Stores in PostgreSQL with pgvector extension

   • Uses Neo4j for graph relationships

   • Employs sentence transformers for semantic embeddings

4. Generates comprehensive report with:

   • Timeline of major developments

   • Technical analysis of competing approaches

   • Assessment of commercial viability

   • Future research directions

Similar to Devin's Notepad, Aesoperator is prompted to open Notes in Ubuntu to write a plan before completing them one by one such that if it fails at some portion of the workflow, it can self heal automatically