Deep Eye

What it is

Deep Eye (deep_eye.py), an MIT-licensed AI-driven pentest CLI. The summary covers its scope; what it doesn't tell you is how you actually stand it up and drive it, and a few operational details worth knowing before you commit.

Preconditions

• Python 3.8+.
• At least one AI provider API key or a local OLLAMA instance — you don't need a paid API to run it.
• Playwright optional (browser tests / CAPTCHA challenge solving); curl_cffi optional (TLS evasion).
• First launch without a config triggers an interactive setup wizard.

What to do

Install:

# Windows
.\scripts\install.ps1        # uses .deep-venv

# Linux/Mac
chmod +x scripts/install.sh && ./scripts/install.sh
source .deep-venv/bin/activate

# Manual
pip install -r requirements-dev.txt
cp config/config.example.yaml config/config.yaml   # then edit API keys

Run and scope scans:

python deep_eye.py -u https://target.com
python deep_eye.py -u https://target.com --formats junit,csv,xlsx
python deep_eye.py -u https://target.com --scope-nl "only /api/* no /logout host target.com"

Diff and retest against a baseline:

python deep_eye.py --diff baseline.json current.json --diff-format html --diff-output diff_report.html
python deep_eye.py -u https://target.com --retest-new baseline.json

CVE intelligence is not bundled live — you build the RAG index yourself:

python scripts/update_cve_database.py
python scripts/build_cve_rag_index.py

Behavior is driven entirely by config/config.yaml (per-provider enabled/model, enabled_checks, payload_generation.context_aware, ai_planner/ai_triage/fp_replay toggles, reporting.dedupe). AI features are gated individually — use_ai, context_aware, and evidence summaries (min_severity: high) are all opt-in flags rather than always-on.

Worth knowing beyond the summary

• 13 providers named in config (openai, claude, grok, ollama, gemini, openrouter, orcarouter, requesty, mistral, groq, lmstudio, litellm, nvidia_nim) — more than the "11" headline; failover routes through a generate() abstraction.
• Export breadth: HTML, PDF, JSON, SARIF, JUnit, CSV, XLSX — SARIF/JUnit make it CI-pipeline-friendly, which the summary doesn't surface.
• CAPTCHA handling (reCAPTCHA, hCaptcha, Cloudflare Turnstile, Arkose) with solver + skip, and a Bug Bounty report writer emitting HackerOne-style Markdown per finding.
• Ships agent skills under .agents/skills/ (pentest, bounty, red/blue, ctf) with a discovery loader in proxy/, plus a pytest suite.

Caveats

Authorized-testing tool only — authors disclaim liability for misuse; unauthorized use is illegal. The page carries a residential-proxy vendor promo (Swiftproxy / InferHub sponsorship); that is advertising, not part of the tool. No version number or release date is given.