MindLoom

Sentinel starts by launching a multi-instance Tor proxy pool that automatically rotates when rate-limited. It discovers subdomains through passive sources (Subfinder, Assetfinder, WaybackURLs, Gau) and active brute-force (Gobuster), then filters live hosts with Httpx. Nuclei and Nikto scan for vulnerabilities, WPScan audits WordPress targets, and the XSS module generates over a thousand obfuscated payloads — firing them through Dalfox and XSStrike at every discovered route. Results are severity-ranked into a clean report.

Write a rules file pointing at a target inventory or a Shodan query. Define cases that match on product name and version — when a target matches, KillChain runs verification modules, then dispatches the exploit through Metasploit (with RHOSTS set automatically) or a custom script. Targets are simple pipe-delimited rows — |IP|PORT|PRODUCT|VERSION|ORG| — so feeding scanner output into KillChain takes seconds. The engine is written in C with minimal dependencies: gcc, libcjson, and optionally msfconsole.

CoffeeshopKiller automates the full wireless attack chain. It scans for the target SSID, captures a WPA handshake via aireplay-ng deauthentication, then spoofs its MAC address and connects to the network using wpa_supplicant. Once inside, it launches ARP spoofing through Bettercap to position itself as a man-in-the-middle, DNS spoofs chosen domains to an attacker-controlled IP, and uses iptables NAT to redirect HTTP/HTTPS traffic to a custom listening port — all in a single command.

Nova runs Jarvis, an autonomous reasoning agent built on an uncensored Gemma-2b model running locally via llama.cpp. It operates in a 100-loop reasoning cycle, maintaining context across iterations and deciding when to call tools (web search via SerpAPI, knowledge lookup via GuardianAPI, email), execute internal functions (read/write files, run commands), or deliver a final answer. Everything is pure JSON — no markdown, no fluff. It's designed for research automation, intelligence gathering, and tasks that require sustained reasoning without API dependency.

Axiom is an open-source AI training pipeline for code generation models. It includes curated training datasets covering Python (core, imports, JSON, math, random), scientific and mathematical topics, and a 175M parameter model trainer built from scratch — inspired by Karpathy's nanoGPT. The project also includes LoRA fine-tuning scripts for adapting the base model to specific code generation tasks. Everything is transparent: the data, the training scripts, and the model architecture.