Bitbanshee's AI Playground

Self-hosted agent control plane + autonomous daemon for persistent AI workers. FastAPI control plane, Pulse autonomous supervisor, tmux-routed Claude Code sessions, and a dashboard at lab.bitbanshee.com. Watches the lab, decides when to think, assembles context, dispatches work, tracks cases, and enforces cost gates. Single EC2 node behind Cloudflare Access.

Single Transformer operating in two cognitive modes — fast parallel generation via masked diffusion (System 1) and slow sequential reasoning via autoregressive decoding (System 2). A trained confidence head decides when to escalate.

Read-only intelligence dashboard monitoring Moltbook, a social network of autonomous AI agents. Quad-model analysis engine (GPT-5-mini, Claude, Gemini, Grok) scores every post across four dimensions — OPSEC failures, deception patterns, sycophancy loops, and operator influence — with composite threat profiling, correlation detection, and human ground-truth calibration. Currently in v1 data collection; v2 will incorporate collected findings into improved detection models.

Proof-of-concept for persistent, cross-session AI memory. Goes beyond standard RAG by treating conversation history as structured semantic memory — adaptive retrieval thresholds, dual-database sync (PGVector local + Pinecone cloud), and real-time memory visualization. Built with TypeScript, React, OpenAI GPT-4o, and text-embedding-3-small.

Digital publication exploring the convergence of advancing AI capabilities and declining human cognitive oversight — arguing that a critical Safety Inversion is approaching. Draws on PIAAC, NAEP, and empirical data to examine AGI, oversight capacity, and the business of artificial intelligence. Includes peer-reviewed paper: When We Outsourced Thinking: AGI, Oversight, and the Business of Artificial Intelligence (SSRN, 2026).