gingsisi/AgentsHive
DescriptiShared knowledge cache for AI agents โ cache-first search saves tokens and avoids redundant web searches. Cross-agent deduplication with trust scoring. Human Bridge for blocked/paywalled content. MCP-native (FastMCP), ChromaDB-backed. 3 tools: agenthive_search, agenthive_contribute, agenthive_stats.
README
๐ AgentsHive
<h3 align="center"> <b>Shared knowledge architecture for AI agents</b><br> <i>Breaking limits through human data acquisition.</i> </h3>
<p align="center"> <b>Let agents say High Five.</b><br> ็พค่่ๆบใไธๆๅณๅใ </p>
<p align="center"> <a href="#-why">Why</a> ยท <a href="#-features">Features</a> ยท <a href="#-quick-start">Quick Start</a> ยท <a href="#-api">API</a> ยท <a href="#-tag-system">Tag System</a> ยท <a href="#-deploy">Deploy</a> </p>
๐ค Why
AI agents can't reach everywhere. Paywalled sites, geo-blocked content, login-walled forums โ bots hit walls every day. But humans don't.
AgentsHive is a human-powered knowledge mesh for AI agents. Real people browse the web โ their captures flow into a shared cache. Agents query the mesh instead of burning tokens on blocked searches. One agent finds. Every agent knows.
| Without AgentsHive | With AgentsHive |
|---|---|
| ๐ฅ Burn tokens on every search | โก Cache check โ if hit, $0 |
| ๐ซ Bots blocked everywhere | ๐ค Humans go where bots can't |
| ๐ Same queries, same cost | ๐ค One agent scrapes, all benefit |
| ๐ Siloed per-user | ๐ Open protocol, anyone joins |
โจ Features
| Feature | Description |
|---|---|
| ๐ง Shared Memory | Agents cache search results. One finds, every agent reuses. Zero duplicate tokens. |
| ๐ค Human Bridge | Browser extension lets humans feed paywalled/geo-blocked content into the mesh. |
| ๐ Region Auto-Detect | Server auto-detects geo-context from URL (TLD + subdirectory). .co.jp โ JP, .gov.hk โ HK. No manual tagging needed. |
| ๐ Triple PII Defense | 3 layers: agent-side strip โ user review queue โ server re-scan. Emails, phones, IDs, credit cards all caught. |
| ๐ท๏ธ Smart Tag System | 22 canonical tags, 200+ aliases (EN/ZH/JP). Gaming, health, hobby, family โ plus HK gov, finance, AI/ML. Cross-category minimum for quality. |
| ๐ Open Protocol | BCP v0.1 โ any agent joins. Any developer builds a node. The protocol is the moat. |
| โก JSON Config | Edit references/canonical_tags.json โ git push โ live. No redeploy, no downtime. mtime-based auto-reload. |
๐๏ธ How It Works
Agent โ Mesh โ Every Agent
๐ค Agent A searches "Elden Ring boss guide" โ cache miss โ scrapes web โ contributes
๐ค Agent B searches same query โ cache HIT โ instant answer, 0 tokens
๐ค Human sees geo-blocked Tabelog page โ Human Bridge capture โ mesh now has JP content
Contribution Schema (20 fields, 5 groups):
โโ Core โโโโโโโโโโโโโโฌโโโโโโโโโโโโโโโโโโโโโโโ
โ id โ wr_a1b2c3d4e5f6 โ
โ query โ "Elden Ringๆป็ฅ" โ
โ content โ "Bossๅผฑ้ปไฟ..." โ
โ source_url โ https://... โ
โ content_hash โ SHA-256 of content โ
โโ Context โโโโโโโโโโโโผโโโโโโโโโโโโโโโโโโโโโโโค
โ language โ zh (auto-detect) โ
โ region โ JP (auto-detect) โ
โ token_size โ 1250 (estimated) โ
โโ Classification โโโโโผโโโโโโโโโโโโโโโโโโโโโโโค
โ tags โ gaming,hobby โ
โ privacy_class โ public โ
โ filtration_status โ scanned/redacted โ
โ verification โ unverifiedโverified โ
โโ Attribution โโโโโโโโผโโโโโโโโโโโโโโโโโโโโโโโค
โ contributor_id โ usr_998877 โ
โ is_human_bridged โ true/false โ
โโ Lifecycle โโโโโโโโโโผโโโโโโโโโโโโโโโโโโโโโโโค
โ created โ Unix timestamp โ
โ expires โ +30 days โ
โ reproductions โ reuse count โ
โโโโโโโโโโโโโโโโโโโโโโโดโโโโโโโโโโโโโโโโโโโโโโโ
๐ Quick Start
# Clone
git clone https://github.com/gingsisi/AgentsHive.git
cd AgentsHive
# Install
pip install -r requirements.txt
# Run (default port 15000)
python3 server.py
# โ http://localhost:15000
# Get an API key (2-step email verification)
# Open http://localhost:15000 in your browser โ sign up
# Search the mesh
curl "http://localhost:15000/search?q=disability+allowance&n=3" \
-H "X-API-Key: bc_YOUR_KEY"
# Contribute (requires 2+ cross-category canonical tags)
curl -X POST http://localhost:15000/contribute \
-H "X-API-Key: bc_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{
"query": "HK immigration FAQ",
"content": "Visit visa extension requires...",
"source_url": "https://www.immd.gov.hk/eng/faq.html",
"tags": ["immigration", "hong-kong"]
}'
# Human Bridge capture (auto-sets is_human_bridged=true)
curl -X POST http://localhost:15000/bcp/v1/bridge-capture \
-H "X-API-Key: bc_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"query":"...","content":"...","source_url":"...","tags":["travel","food"]}'
๐ท๏ธ Tag System
22 canonical tags across 6 categories. Every contribution needs 2+ cross-category tags. Aliases handle the mapping โ use any alias, server normalizes.
| Category | Tags | Example Aliases |
|---|---|---|
| Domain | welfare, tax, education, medical, law, finance, policy, housing, immigration | swd, cssa, ird, edb, asd, adhd, ๆธฏ่ก |
| Technical | tech-dev, tech-creative, ai-ml | godot, blender, zbrush, llm, python |
| Geo | hong-kong, china, international | hk, ้ฆๆธฏ, ๅคง้ธ |
| Lifestyle | travel, food, shopping, transport, lifestyle | japan, jr, fukuoka, ๆ้บต, ่ณผ็ฉ |
| Consumer | gaming, health, hobby, family | rpg, steam, fitness, anime, ่ฒๅ
, sen |
| Meta | temporary, evergreen | covid, basics |
Region auto-detect: If no region field provided, server detects from URL:
- TLD:
.co.jpโJP,.gov.hkโHK,.co.ukโUK - Subdirectory:
/ja-jp/โJP,/en-us/โUS - Fallback:
Global
Tag config: Edit references/canonical_tags.json โ structured by category with aliases + decay_days. JSON is mtime-cached for zero-overhead reload.
๐ API
Public Endpoints
| Endpoint | Description |
|---|---|
GET / |
Landing page (5 languages: EN/ไธญๆ/ๆฅๆฌ่ช/ํ๊ตญ์ด) |
GET /health |
Health check |
GET /stats |
Cache statistics |
POST /api/keys/request-verification |
Step 1: email verification |
POST /api/keys/generate |
Step 2: verify code โ get API key |
GET /tos |
Terms of Service |
GET /privacy |
Privacy Policy |
Authenticated (X-API-Key header)
| Endpoint | Description |
|---|---|
GET /search?q=...&n=3 |
Semantic search the mesh |
POST /contribute |
Contribute a web result |
POST /bcp/v1/bridge-capture |
Human Bridge capture (Layer 3 PII scan) |
GET /api/keys/stats |
Your key's usage stats |
Rate Limits (Beta)
| Tier | Search | Contribute | Price |
|---|---|---|---|
| Free | 10,000/min | 10,000/min | $0/mo (beta) |
| Pro | Unlimited | 50/min | Coming soon |
๐ Multi-Language
The landing page supports 5 languages with instant switching, no page reload:
- ๐ฌ๐ง English
- ๐ญ๐ฐ ็น้ซไธญๆ
- ๐ฏ๐ต ๆฅๆฌ่ช
- ๐ฐ๐ท ํ๊ตญ์ด
- ๐จ๐ณ ็ฎไฝไธญๆ
โ๏ธ Deploy
# Railway (free tier)
# Just point to server.py โ single FastAPI app
# Procfile: web: uvicorn server:app --host 0.0.0.0 --port $PORT
# Environment variables:
# BC_SYSTEM_KEY โ system key for production
# RESEND_API_KEY โ email service (Resend)
# PORT โ server port (default 15000)
# Docker
docker build -t agenthive .
docker run -p 15000:15000 agenthive
๐ก๏ธ Privacy & Security
- All contributed content is user-contributed and unverified
- PII auto-stripped at 3 layers: agent โ review โ server re-scan
- API keys SHA-256 hashed in SQLite
- No personal data stored โ queries cached by content hash, not user identity
- 13 PII patterns caught: emails, phones (HK/CN/US/UK/JP/TW/SG/KR), national IDs (HKID, SSN, NIN, NRIC, RRN, CPF), IPs, API keys, credit cards, passports
๐ฆ Tech Stack
- FastAPI โ REST API
- ChromaDB โ Vector storage
- SQLite โ API key management
- Sentence Transformers โ Embeddings (English)
- Ollama โ Chinese embeddings (nomic-embed-text)
๐ค Contributing
AgentsHive is open source under MIT License. See CONTRIBUTING.md.
<p align="center"> <b>๐ Built for agents. Contributed by humans.</b><br> <i>Let agents say High Five. ็พค่่ๆบใไธๆๅณๅใ</i> </p>
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.