Artificial IntelligenceCompleted

Autonomous Multi-Source Research Agent

A self-directed AI research agent built from scratch using ReAct planning loops, grounded web search, and vector synthesis.

TypeScriptLLMsVector DBReActNode.js
GitHub RepositoryLive Demo

Autonomous Multi-Source Research Agent

An open-source, modular research agent designed to demonstrate function calling, planning loops, and grounded knowledge retrieval without high-level black-box frameworks.

Architecture Overview

User Topic Request
       │
       ▼
[ ReAct Agent Loop ] ──► (1. Reason) ──► Determine missing knowledge
       │
       ├──► (2. Tool Call) ──► Query Web Search / Local Vector Store
       │
       ├──► (3. Observe)   ──► Ingest API response / document chunk
       │
       └──► (4. Synthesize) ──► Produce final Markdown report

Key Features

  1. Deterministic Function Schemas: Strictly typed tool definitions for search, webpage parsing, and file output.
  2. Infinite Loop Safeguards: Token counters, maximum iteration limits, and timeout boundaries.
  3. Structured Citation Generation: Embeds source links and document metadata in final Markdown outputs.