BeHive — Open-Source Deep Research Engine

v0.1.0 — Open Source (MIT)

Structuredknowledge,

nottextsoup.

Feed it a question. Get back scored claims,entity graphs, and asynthesized report — from 1000+ live sources.

Research ▋

$pip install behive

View on GitHub

behive research "NVIDIA GPU market 2026" --scale 30

BeHive CLI Demo — animated terminal showing research pipeline with bee flight physics, honeycomb progress bars, and knowledge graph output

API Sources

Endpoints

0.0

Avg Quality

API Endpoints

⬡ MCP Native⬡ 70+ APIs⬡ Knowledge Graph⬡ Quality Scoring⬡ Self-Hosted⬡ MIT License⬡ SSE Streaming⬡ Stealth Drones⬡ Neo4j⬡ PostgreSQL⬡ Docker Compose⬡ MCP Native⬡ 70+ APIs⬡ Knowledge Graph⬡ Quality Scoring⬡ Self-Hosted⬡ MIT License⬡ SSE Streaming⬡ Stealth Drones⬡ Neo4j⬡ PostgreSQL⬡ Docker Compose⬡ MCP Native⬡ 70+ APIs⬡ Knowledge Graph⬡ Quality Scoring⬡ Self-Hosted⬡ MIT License⬡ SSE Streaming⬡ Stealth Drones⬡ Neo4j⬡ PostgreSQL⬡ Docker Compose

⬢ MCP Native⬢ 70+ APIs⬢ Knowledge Graph⬢ Quality Scoring⬢ Self-Hosted⬢ MIT License⬢ SSE Streaming⬢ Stealth Drones⬢ Neo4j⬢ PostgreSQL⬢ Docker Compose⬢ MCP Native⬢ 70+ APIs⬢ Knowledge Graph⬢ Quality Scoring⬢ Self-Hosted⬢ MIT License⬢ SSE Streaming⬢ Stealth Drones⬢ Neo4j⬢ PostgreSQL⬢ Docker Compose⬢ MCP Native⬢ 70+ APIs⬢ Knowledge Graph⬢ Quality Scoring⬢ Self-Hosted⬢ MIT License⬢ SSE Streaming⬢ Stealth Drones⬢ Neo4j⬢ PostgreSQL⬢ Docker Compose

The Problem

You ask AI to research a topic.
It gives you confident-sounding
text soup.

✗

Without BeHive

"Based on my training data, NVIDIA's GPU market shows promising growth with several key factors driving demand..."

No sourcesNo structureStale data

✓

With BeHive

[0.94] NVIDIA H200 production at 500K units/month as of Q2 2026

[0.91] Data center GPU revenue $35.1B in FY2026 Q2, 87% YoY

[0.88] Blackwell GB200 yield rate: 92% at TSMC CoWoS-L

290 claims234 sources8 min

How It Works

Swarm intelligence.
Not prompt engineering.

🐝Step 01

Scout

Queen decomposes your question into 5 research axes. Generates 60+ search queries. Checks 70+ APIs. Plans the mission.

⬡Step 02

Harvest

Stealth drones fetch 1000+ URLs through an 8-layer evasion stack. Bypasses Cloudflare, DataDome, Akamai. No IP gets burned.

⚗️Step 03

Extract

Dual-model extraction. Haiku for bulk speed. Sonnet for enriching thin claims. Every fact scored 0.0–1.0 on 5 dimensions.

🧬Step 04

Deduplicate

Jaccard similarity at 0.60 threshold merges duplicate facts from different sources. No inflation. Only unique intelligence.

📊Step 05

Synthesize

Claude weaves verified claims into a structured report. Every statement mapped to a scored source. Zero hallucination.

🕸️Step 06

Graph

Entities and relationships stored in Neo4j. Knowledge compounds across missions. Research builds on research.

Benchmarks

Real results. No cherry-picking.

Scale 30 (standard). EC2 g6.24xlarge — 4× NVIDIA L4, 96 vCPU, 384 GB RAM.

Topic	Claims	Quality	Duration	Sources
NVIDIA GPU market 2026	290	0.797	8 min	234
OpenAI GPT-5 capabilities	574	0.789	12 min	174
EU AI Act enforcement	267	0.759	6 min	130
Meta Llama 4 architecture	568	0.821	11 min	198
SpaceX Starship progress	412	0.804	9 min	167

Honest scoring. No sigmoid rescaling. No artificial inflation.
Weighted average of specificity, density, uniqueness, verifiability, and structure.

Integrations

Works with your existing stack.

🤖

Claude Desktop

MCP (zero-code)

Add one URL to config. Claude calls BeHive natively.

⌨️

Cursor / Windsurf

MCP

Research from your IDE. No context switching.

⚡

n8n

Community Node

Drag-and-drop BeHive into automation workflows.

🏛️

Hermes Agent

MCP + Skill

Deep research with autonomous agents.

🦞

OpenClaw

Skill

ClawHub-ready skill for OpenClaw agents.

🔌

Any MCP Client

Streamable HTTP

Standard MCP protocol. Any client works.

MCP Configuration (works everywhere)

{
  "mcpServers": {
    "behive": {
      "url": "http://localhost:8090/mcp",
      "transport": "streamable-http"
    }
  }
}

Self-Hosting

Three commands. Your infrastructure.

Docker Compose. PostgreSQL + Neo4j + API. Zero config.

bash

$ git clone https://github.com/qa10devteam/behive.git
$ cp .env.example .env  # add your API key
$ docker compose up -d

✓ API ready at localhost:8091
✓ MCP server at localhost:8090
✓ PostgreSQL + Neo4j running

Comparison

Why not just use Perplexity?

	BeHive	GPT-Researcher	Tavily	Perplexity
Per-claim scoring	✅	—	—	—
Knowledge graph	✅	—	—	—
MCP native	✅	—	—	—
70+ API sources	✅	—	~	~
Self-hosted	✅	~	—	—
SSE streaming	✅	—	—	—
Cross-session KG	✅	—	—	—
Price	Free	Free	$29/mo	$20/mo+

Start researching.

One command. Infinite knowledge.