Multi-Agent AI Research System

Research any topic with 10+ AI Agents

Watch AI agents collaborate in real-time to research, fact-check, and synthesize comprehensive reports with full citation graphs and trust scores.

10+
AI Agents
8+
Search Tools
5-dim
Trust Score
Real-time
SSE Stream
Deep Research Swarm — Sample Run Stats
14
Agents Spawned
23
LLM Calls
47
Tool Calls
12
Claims Verified

Everything you need

A production-grade multi-agent research system with real-time observability.

Multi-Agent Research

AI agents collaborate to research any topic — Searchers, Browsers, Fact-Checkers, and Critics working in parallel.

Real-Time Dashboard

Watch agents work live with react-flow visualization. See every tool call, status change, and citation as it happens.

Trust Scoring

Every claim is independently verified with 5-dimension trust scoring: authority, agreement, recency, fact-check, and internal consistency.

Citation Graph

Neo4j-powered knowledge graph connecting claims, sources, and sub-questions. Click any node to explore the evidence chain.

Intelligent Routing

Planner assigns optimal tools per sub-question: arXiv for papers, GitHub for code, Serper/Tavily for news, pgvector for internal docs.

Built with LangGraph

Orchestrator-worker pattern with stateless async agents, slot-based write isolation, and circuit breakers.

How it works

Orchestrator-worker pattern with stateless agents and shared memory.

Agent Pipeline — Parallel Execution Flow

User Query
The research question to investigate
Planner
Hypothesis tree + search strategy
Lead Orchestrator
Decomposes into 3–8 sub-questions + assigns tools
Searcher Agents
3–8 parallel workers (one per sub-question)
Sub-question → Search
Query normalization
Multi-tool synthesis
Browser Workers
2–4 parallel workers
Deep URL fetch
JS rendering
Content extraction
Critic
Gap analysis + loop decision (max 2 rounds)
Find gaps
Contradictions
Loop or proceed
Fact-Checker Agents
Parallel claim verification
Claim extraction
Independent search
Trust score (5-dim)
Synthesizer
Integrates all findings → coherent report
Citation Formatter
Embeds verified citations into the final report
Final Report
Claims + Trust scores + Citations

Tech Stack

Python
Core backend logic
TypeScript
Type-safe full-stack
Next.js
React dashboard
FastAPI
Backend framework with SSE streaming
LangGraph
Multi-agent orchestration
react-flow
Agent graph visualization
Neo4j
Citation knowledge graph
PostgreSQL
State store + pgvector
Redis
Hot state cache
Docker
Containerized deployment
Langfuse
LLM observability & tracing
OpenTelemetry
System traces & metrics

Tools

Tavily
LLM-optimized web content with advanced search depth
Serper
Google Search API for maximum freshness
Exa.ai
Semantic-first search, ideal for academic content
arXiv
Structured metadata for research papers and preprints
GitHub
Repo search, README extraction, code analysis
DuckDuckGo
Unlimited free fallback search — never runs out
Playwright
JS-rendered pages, PDFs, paywalled previews
pgvector
User-uploaded document search via vector similarity

Ready to see it in action?

Sign in and start researching. Watch agents collaborate in real-time.