Ryan Bell (solo)

AirStack — Know before you renew: dated BUILD/BUY/KEEP/RETIRE verdicts for AI tools, with evidence that expires. Software advice expires. A single model or platform release can absorb a tool's core capability overnight, yet teams renew on year-old judgments. AirStack keeps each tool decision current. A fleet of Muse Spark agents builds an append-only evidence graph from official, keyless sources: GitHub changelog and releases, Google Workspace updates, OpenRouter, Vercel AI Gateway and Hugging Face. From that graph it issues a verdict for each tool: BUILD, BUY, KEEP or RETIRE. Each verdict comes with its decisive reason, the evidence IDs behind it and a "last verified" time. Every verdict carries its expiry condition as a saved watch. When a new fact matches a watch, the verdict is withdrawn (LAPSED) before any reassessment. The agents then publish a supported revision or leave an honest unresolved state. "Not RETIRE" never defaults to KEEP, and stale advice is never served silently. Watch matching runs in code first: no key match, no model call. That keeps inference near zero unless the world actually changes. The same evaluate() function powers public verdicts, erosion pages for famous suites, and seat-by-seat reviews of an organization's own stack. Example: when Google announced in-person meeting notes on Aug 13, an Android, English-speaking Workspace Business Standard seat can retire its note-taker. A Hindi+English seat keeps it, because Hindi isn't supported. An iPhone seat waits for the Aug 31 iOS rollout. These outcomes come from the inputs, not from scripted personas. Everything is also exposed as an MCP server. In the demo, a brand-new Muse agent that knows nothing about AirStack calls the tools and answers "should we adopt X?" with evidence it can audit. Built solo by a founder who faces this build/buy/keep/retire call every week. Live demo: https://airstack.bot