Private Deep Search

Performs private, multi-round deep web searches excluding Google/Bing, synthesizes results with citations, and does not retain user data or logs.

Security Scan

The skill's code, instructions, and requirements are consistent with a local, privacy-focused deep web search assistant — nothing in the bundle appears to be doing unrelated or covert actions.

Purpose & Capability

Name and artifacts match: this is a self-hosted SearXNG-based search + a Python 'deep_research' tool that iteratively queries SearXNG and scrapes pages. Required software (Docker, Python, aiohttp, BeautifulSoup) is appropriate for the stated task; there are no unrelated credentials, binaries, or config paths requested.

Instruction Scope

Runtime instructions (setup.sh, docker-compose, SKILL.md) remain within the expected scope: start a local SearXNG container, copy skills into ~/.clawdbot/skills, and run deep_research.py which fetches and scrapes web pages returned by SearXNG. One operational note: the Python tool will fetch arbitrary result URLs (intended), so it will make outbound HTTP requests to the internet (and could reach any host that appears in search results). This is expected for a scraper but is a network-privacy consideration rather than incoherence.

Install Mechanism

There is no complex/install spec; setup uses docker-compose to pull searxng/searxng:latest (official Docker Hub image) and a local setup script to generate a secret key. Pulling an image from Docker Hub is normal here, but using the mutable 'latest' tag has supply-chain implications (image contents can change). Python deps are standard and installed via pip when requested.

Credentials

The skill does not request environment variables, credentials, or system config paths. It asks for reasonable local dependencies (Docker, Python and a couple of Python packages) which match the described functionality.

Persistence & Privilege

Skill is not marked always:true and does not require modifying other skills or system-wide agent settings. setup.sh updates only the included settings.yml and starts the local container; the container mounts only the local ./searxng config directory. No excessive privileges are requested in the manifest.

Assessment

This package appears coherent and implements a local SearXNG + scraper. Before installing: 1) Inspect the Docker image (searxng/searxng:latest) or pin a specific release tag to avoid surprise image changes. 2) Verify docker/searxng/settings.yml after setup (setup.sh replaces the placeholder secret for you). 3) Be aware deep_research.py will fetch and parse URLs returned by searches — consider running the container and scraper in an isolated network namespace or behind a VPN if you want to limit outbound reach (prevents accidental access to internal hosts). 4) Review and, if necessary, adjust IGNORED_DOMAINS and request timeouts/rate limits to avoid aggressive scraping. 5) If you require extra assurance, pull and inspect the searxng Docker image locally or build from source and run with least-privilege container settings (no host network unless you intend VPN=host). Overall: functionally consistent with its stated purpose, but follow standard supply-chain and network-isolation best practices.

privatedeepsearch-melt

"Google wants to know everything about you. I want to know nothing." — melt, probably

melt is your privacy-obsessed research assistant. She searches the web deeply, synthesizes findings with citations, and forgets everything the moment she's done.

Powered by SearXNG. No Google. No tracking. No API keys. No BS.

---

What melt Does

🔍 She Searches (Privately)

searx "best password managers 2026" 5

melt queries DuckDuckGo, Brave, Startpage, and friends. Google and Bing are blocked at the door.

🔬 She Researches (Deeply)

deep-research "zero knowledge proofs practical applications"

melt doesn't just search once and call it a day. She:

1. Searches your query
2. Reads the results
3. Thinks "hmm, I need more context"
4. Searches again with refined terms
5. Scrapes full article content
6. Repeats up to 5 times
7. Synthesizes everything into a report with citations

Like Perplexity, but she doesn't sell your soul to investors.

🛡️ She Protects (Always)

What Big Tech Does	What melt Does
Logs every search	Logs nothing
Builds a profile on you	Forgets you exist
Sells your data	Has no data to sell
Runs on their servers	Runs on YOUR machine
Costs $20/month	Costs $0/forever

---

Quick Start

1. Wake melt Up

# Auto-setup (generates secret key + starts container)
./setup.sh

# Or manually
cd docker && docker-compose up -d

She'll be ready at http://localhost:8888

2. Teach Her to Your AI

cp -r skills/* ~/.clawdbot/skills/

# Or via ClawdHub
clawdhub install privatedeepsearch-melt

3. (Optional) Fire the Competition

Tell Clawdbot to stop using Brave API:

{
  "tools": {
    "web": {
      "search": { "enabled": false }
    }
  }
}

---

How Deep Research Actually Works

You: "explain quantum computing"
                    │
                    ▼
    ┌───────────────────────────────┐
    │  melt: "Got it. Let me dig."  │
    └───────────────┬───────────────┘
                    │
    Round 1: "explain quantum computing"
    Round 2: "quantum computing detailed analysis"
    Round 3: "quantum computing comprehensive guide"
                    │
                    ▼
    ┌───────────────────────────────┐
    │  SearXNG: *queries 5 engines* │
    │  Returns 10 results per round │
    └───────────────┬───────────────┘
                    │
                    ▼
    ┌───────────────────────────────┐
    │  melt: "YouTube? Facebook?    │
    │         Nice try. BLOCKED."   │
    └───────────────┬───────────────┘
                    │
                    ▼
    ┌───────────────────────────────┐
    │  *Scrapes 10 pages at once*   │
    │  asyncio go brrrrrr           │
    └───────────────┬───────────────┘
                    │
                    ▼
    ┌───────────────────────────────┐
    │  # Deep Research Report       │
    │  **Sources:** 17              │
    │  ## [1] Quantum 101...        │
    │  ## [2] IBM's Breakthrough... │
    └───────────────────────────────┘

---

Privacy Architecture

Your brain
    │
    ▼ (you type a query)
┌─────────────────┐
│   Clawdbot      │  ← Your machine. Your rules.
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│     melt        │  ← Localhost. No cloud. No logs.
│   (SearXNG)     │
└────────┬────────┘
         │
         ▼ (optional but recommended)
┌─────────────────┐
│    Your VPN     │  ← Hide your IP from everyone
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│  DuckDuckGo     │  ← They see VPN IP, not you
│  Brave Search   │
│  Startpage      │
└─────────────────┘

Who sees what:

• Google: Nothing. Blocked.
• Your ISP: Encrypted traffic. They mad.
• melt: Everything. But she has amnesia.

---

Why Open Source Matters

melt is MIT licensed because:

1. You can audit the code — No hidden trackers
2. You can fork it — Make your own version
3. You can improve it — PRs welcome
4. You own your data — It never leaves your machine

Closed-source "privacy" tools ask you to trust them. melt asks you to verify.

---

Engines melt Trusts

✅ Enabled:

• DuckDuckGo, Brave Search, Startpage
• Qwant, Mojeek
• Wikipedia, GitHub, StackOverflow, Reddit, arXiv
• Piped, Invidious (YouTube without YouTube)

❌ Blocked:

• Google (all of it)
• Bing (all of it)
• Anything that tracks you

---

Requirements

• Docker & Docker Compose
• Python 3.8+
• A healthy distrust of Big Tech

pip install aiohttp beautifulsoup4

---

Files

privatedeepsearch-melt/
├── README.md              ← You are here
├── docker/
│   ├── docker-compose.yml ← SearXNG deployment
│   └── searxng/settings.yml
├── skills/
│   ├── searxng/           ← Basic search skill
│   └── deep-research/     ← The good stuff
└── docs/
    ├── PRIVACY.md         ← How melt protects you
    └── TROUBLESHOOTING.md ← When things break

---

Credits

• SearXNG — The real hero. Privacy-respecting meta-search that makes this possible.
• OpenWebUI Deep Research — Algorithm inspiration
• Clawdbot — AI assistant framework

---

License

MIT — Do whatever you want. Just don't be evil.

---

"The best search history is no search history." — melt

SearXNG 🛡️