behive-research

BeHive Deep Research

Overview

BeHive is a deep research engine that extracts structured, scored claims from any topic. Unlike simple web search, it:

1. Decomposes topics into research axes
2. Fetches 1000+ sources via 8-layer stealth drones (70+ APIs)
3. Extracts typed claims with per-claim quality scores (0.0–1.0)
4. Builds entity relationship graphs (Neo4j)
5. Produces synthesized reports with inline citations

This skill connects Hermes Agent to a running BeHive instance.

When to Use

• User says "research X deeply", "gather intelligence on Y", "find everything about Z"
• User needs verified facts with sources, not LLM-generated summaries
• User wants claims they can cite, filter by quality, or query later
• User asks to build a knowledge base on a topic over time
• User says "run a mission on X", "scale 100", "deep dive"

Don't Use For

• Simple factual lookups (use web_search)
• Single-page content extraction (use web_extract)
• Real-time news (BeHive takes 5-15 min per mission)

Prerequisites

BeHive API running at http://localhost:8091 or configured endpoint.

Verify:

curl -s http://localhost:8091/health | jq .

If using MCP (recommended), configure in ~/.hermes/config.yaml:

mcp_servers:
  behive:
    url: http://localhost:8090/mcp
    transport: streamable-http

Quick Research (MCP)

If BeHive MCP is configured, use MCP tools directly:

mcp_behive_research_topic(request={"query": "NVIDIA GPU market 2025-2026", "depth": 3, "force": true})

Then poll:

mcp_behive_mission_status(job_id="<returned_id>")

Get report:

mcp_behive_get_report(job_id="<id>", format="markdown")

Search past knowledge:

mcp_behive_search_knowledge(query="NVIDIA revenue", limit=20)

Research via REST API (terminal)

Start a Mission

curl -s -X POST http://localhost:8091/research \
  -H "Content-Type: application/json" \
  -d '{"topic": "EU AI Act enforcement mechanisms 2026", "scale": 30, "depth": 3}' | jq .