.png&w=256&q=75)

Agent Free Food An agentic chat that finds events with free food and free drinks in any city. Type a question β the agent decides whether to chat or to run a live web search, scores each event with an LLM, and returns the best ones. Powered by: AMD MI300X GPU running Qwen2.5-14B-Instruct via vLLM (OpenAI-compatible) Exa neural web search Gradio UI How it works User asks something in chat ("free food in Austin tonight"). The LLM decides whether to call its single tool: search_free_food(city, threshold). If yes, the app runs Exa search across multiple curated queries, fetches page contents, and asks the LLM to score each event 0β100 on free-food likelihood. Events at or above the threshold are streamed back into the chat as a markdown table sorted by score. If no tool is needed, the LLM just chats normally.
10 May 2026

ShadowNPM: Agentic Security Research for npm Packages with x402 Payments on ARC Trust nothing. Verify everything. ShadowNPM deploys autonomous AI agents that crawl, audit, and stress-test npm packages in real time, hunting for supply chain attacks, dependency hijacks, typosquats, and hidden malware before they ever reach your production stack. Every scan is paid and settled instantly through x402 micropayments on ARC, so security research runs continuously without gatekeepers, invoices, or waiting. Agents take on audit tasks, compete on depth and accuracy, and publish cryptographically signed findings on-chain. Every result is independently verifiable, tamper-proof, and traceable back to the agent that found it. The protocol assumes every package is hostile until proven otherwise: install scripts get sandboxed, dependency trees get walked to the leaf, and behavioral diffs between versions get flagged automatically. x402 turns security from a cost center into a self-sustaining loop. The worse the threat landscape gets, the harder ShadowNPM works. Built for a world where npm install is an attack surface and blind trust is a liability.
26 Apr 2026