Open Source & Experiments

Session handoff tool built for AI agent workflows. POST a markdown briefing, get a shareable URL. Content is first-write-wins (no auth needed), auto-expires, and serves raw markdown to CLI tools while browsers get a rendered SPA with expiry countdown and view counts. Two source files, no build step, deployed on Cloudflare Workers.

The cockpit for your copilots. VibeCockpit auto-detects installed AI coding tools and gives you one place to search, filter, and resume any session. Supports Claude Code, Claude Desktop, Codex CLI, Copilot CLI, OpenCode, and Gemini CLI. Web dashboard, terminal TUI, remote SSH scanning, and a plugin architecture for adding new tools. Single binary, zero config, open source.

eBPF-based auditor for CI/CD pipelines that captures environment variable access via libc getenv to detect secret exfiltration. Records DNS queries (UDP port 53) to track external connections. Outputs JSON audit reports with timestamps, process names, and PIDs/TIDs. Includes .env.monitor allowlist for targeting specific secrets. Runs as a GitHub Actions step or on any Linux host with eBPF support.

A proof-of-concept implementation of the AI firewall concept from the blog series. Sits between clients and Ollama, inspects every prompt via a secondary inspector LLM, and blocks requests above a configurable risk threshold. Includes a web dashboard with per-request timing, color-coded risk levels, and configurable inspector prompts (standard, strict, multilingual, custom).

A public-facing threat intelligence feed built at VMRay. Aggregates and serves real-time malware analysis data from VMRay’s sandbox infrastructure. Built with serverless architecture on AWS.