Skip to content

SKILL.md

Latest commit

 

History

History
236 lines (174 loc) · 7.49 KB

SKILL.md

File metadata and controls

236 lines (174 loc) · 7.49 KB
Raw
name agent-shielded
description Agent Shielded — Agent Privacy Cloud. Private infrastructure for AI agents in one command: Ollama, n8n, Qdrant, Agent Sandbox, Tor Rotator, Privoxy, Gotify, Agent Dashboard + Prometheus, Grafana, Portainer. Zero data leaks. Triggers: 'agent shielded', 'agent privacy', 'AI agent infrastructure', 'private LLM', 'Ollama setup', 'n8n workflows', 'vector database', 'agent sandbox', 'open source AI', 'private agent', 'agent infrastructure', 'freedom stack agents'.

Skill Agent Shielded — Freedom Stack Agent Privacy Cloud

1 command. Zero data leaks. Complete private infrastructure for AI agents.

Repo: https://github.com/Michae2xl/freedom-stack

What This Skill Does

Installs the full Agent Privacy Cloud at once:

ssh root@YOUR_VPS_IP
curl -fsSL https://raw.githubusercontent.com/Michae2xl/freedom-stack/main/scripts/install.sh -o install.sh
bash install.sh --agents --searxng --tor --security --domain yourdomain.com

# No domain (works via IP + Tor .onion):
bash install.sh --agents --searxng --tor --security

Installed Services

Agent Privacy Cloud (8 components)

Service Port What It Does
Ollama + Open WebUI :11434 / :8181 Local LLM — zero data to OpenAI/Anthropic
n8n :5678 Visual AI workflow orchestration
Qdrant :6333 Vector DB — long-term semantic memory
Agent Sandbox — Isolated Python 3.12 + Node 20
Tor Rotator :9060 New circuit every 30s for scraping without bans
Privoxy :8118 HTTP proxy via Tor — anonymous requests
Gotify :8070 Private push notifications to phone
Agent Dashboard :3333 Real-time status of all infra

Infrastructure

Service Port Function
Prometheus :9090 Metrics
Grafana :3100 Dashboards
Portainer :9000 Docker GUI
SearXNG :8888 Anonymous search (JSON API enabled)

Agent Flow

Agent (sandbox)
    |---> SearXNG      :8888  anonymous search — GET /search?q=QUERY&format=json
    |---> Ollama       :11434 local LLM — POST /api/generate
    |---> Qdrant       :6333  vector memory — PUT /collections/memory/points
    |---> Tor/Privoxy  :9060/:8118  anonymous network for external requests
    |---> Gotify       :8070  push notifications to owner
    |---> n8n          :5678  complex workflow orchestration

Code example (inside the sandbox):

import requests

# 1. Anonymous search (Google/Bing/DDG never know who you are)
results = requests.get(
    "http://searxng:8080/search",
    params={"q": "bitcoin price", "format": "json"}
).json()

# 2. Analyze with local LLM (zero data to cloud)
analysis = requests.post("http://ollama:11434/api/generate", json={
    "model": "llama3.2:3b",
    "prompt": f"Analyze: {results['results'][0]['content']}",
    "stream": False
}).json()["response"]

# 3. Store vector memory (persistent)
requests.put("http://qdrant:6333/collections/memory/points", json={
    "points": [{"id": 1, "vector": [0.1]*384, "payload": {"analysis": analysis}}]
})

# 4. Anonymous external request via Tor
data = requests.get(
    "https://api.example.com/data",
    proxies={"https": "socks5h://tor-rotator:9050"}
).json()

# 5. Notify owner on phone
requests.post("http://gotify:80/message", json={
    "title": "Agent Alert", "message": analysis[:100], "priority": 5
})

Use Cases

1. Trading / Betting Agents

Agent searches odds via SearXNG → analyzes with Ollama → executes via Tor (anonymous) → stores in Qdrant → notifies via Gotify.

2. OSINT / Research

Scraping via Tor Rotator (new circuit every 30s, no IP bans) → findings stored in Qdrant as embeddings → semantic cross-reference.

3. Agent Swarms

50+ agents in sandbox → communication via Matrix E2E → shared memory in Qdrant → orchestration via n8n visual workflows.

4. Enterprise Private AI

Local LLM (Ollama, zero cloud) → GDPR-compliant storage → HIPAA without cloud vendors.

5. Investigative Journalism AI

Agent monitors sources via Tor → cross-references data → stores encrypted evidence → communicates via Matrix E2E + .onion.

How To Use This Skill

When the user arrives, ask:

  1. Goal? (what type of agent do you want to run?)
  2. Have a VPS? (specs: 16GB+ RAM recommended for Ollama)
  3. Have a domain? (optional — works via IP/Tor .onion)
  4. Want human services too? (if yes, use the sovereign skill alongside)

Based on answers:

Situation Command
Full Agent Cloud --agents --searxng --tor --security
Agent Cloud + domain --agents --searxng --tor --security --domain X
Agent Cloud + human services --all --domain X (installs everything)
Ollama + n8n only (minimal) --agents

Requirements

Minimum Recommended
RAM 8GB (without Ollama) 16GB+ (with Ollama)
CPU 2 vCPUs 4+ vCPUs
Disk 40GB 100GB+ (LLM models: 2-7GB each)
OS Ubuntu 22.04 Ubuntu 24.04
Cost ~EUR 8/mo ~EUR 18/mo (Hetzner CX32)

Recommended Ollama Models

# Light (2GB RAM, fast)
docker exec freedom-ollama ollama pull llama3.2:3b

# Balanced (5GB RAM)
docker exec freedom-ollama ollama pull llama3.1:8b

# Powerful (12GB RAM)
docker exec freedom-ollama ollama pull llama3.1:70b

Privacy-Friendly VPS Providers

Provider Privacy Price (16GB)
Hetzner High, GDPR EUR 18/mo
Njalla Maximum, Zero KYC ~EUR 30/mo
1984.is Maximum, Iceland ~EUR 25/mo

Troubleshooting

# Check status
docker compose ps

# Service logs
docker compose logs freedom-ollama --tail 50

# Test Tor
curl --socks5-hostname 127.0.0.1:9060 https://check.torproject.org/api/ip

# Test SearXNG JSON
curl "http://localhost:8888/search?q=test&format=json"

# Test Ollama
curl http://localhost:11434/api/generate -d '{"model":"llama3.2:3b","prompt":"hello","stream":false}'

# Free memory for Ollama
echo 3 > /proc/sys/vm/drop_caches

# Full diagnostic script
bash scripts/troubleshoot.sh

Known bugs and fixes (v4.0)

  • Stalwart Mail: correct image is stalwartlabs/stalwart:latest
  • Agent Dashboard: correct npm package is dockerode (not docker-dockerode)
  • Privoxy: does not depend on Tor — works independently
  • SearXNG JSON: formats: [html, json] required in settings.yml
  • Ollama out of memory: run echo 3 > /proc/sys/vm/drop_caches first

Market Position

                    PRIVACY
                         |
                         |   * AGENT SHIELDED (Freedom Stack)
                         |     (privacy-native + agents)
                         |
            Njalla/1984  |
            (VPS only)   |    n8n AI Kit, Dify, LangChain
                         |    (self-hosted, no real privacy)
          ---------------+------------------------> AI AGENTS
                         |
                         |    OpenAI / AWS / Google Cloud
                         |    (cloud, logs everything)

The only product in the high-privacy + AI agent infrastructure quadrant.

Relation to Sovereign Stack

Project Audience Focus
Agent Shielded / Freedom Stack (this skill) Devs/AI Agents Agent Privacy Cloud
Sovereign Stack (sovereign skill) Humans Personal privacy, degoogle

They can run together on the same VPS — fully complementary.