recently liberated from n5. they let me go right before launching (and rapidly un-launching) a product that my due diligence flagged as a regulatory time bomb. i’m taking that as a sign to keep my seven-day workweeks focused on building things that don’t violate federal law — specifically pushing forward on brane and destroysaas.coop.

keeping receipts. nickel-5 is a private archive of recordings and analysis from my time at Nickel5, Inc. — for regulatory and civil filings. no commentary here; the documents speak for themselves.

shipping:

• xenv — single-binary secrets manager for ai coding agents. AES-256-GCM vaults safe to commit, decrypted secrets never touch disk, MCP server baked in.
• echolat — native mobile SSH terminal for babysitting AI agents in remote tmux sessions from my phone (private, soon).
• talon — mobile agent bridge. agents drive a real phone’s webview over a relay to bypass datacenter fingerprinting and 2FA walls — your pocket is the residential IP (private, in PRD).
• a16zee — the cocaine run. parody of a16z’s speedrun, for people who actually ship. #vcevil.
• VRPS api — bun.js + supabase parking platform, clean IA, replacing the rails monolith. also: the people at VRPS are some of the kindest humans i’ve ever worked with 🫶 (private).
• joust — adversarial architecture compiler. LLMs don’t agree, so make them fight about it. agent-installable via claude skill.
• spool — filesystem convention for serial multi-session work via linked artifacts. agents and teams, same protocol.
• VRPS mon — infrastructure status dashboard. panopticon for the parking platform (private).
• VRPS lpr-rails — the legacy LPR rails app i’m slowly euthanizing in favor of the new stack (private).
• bunny — dark factory. ai agents build software autonomously. claude implements, gemini hardens, humans intervene only when stuck.
• gh-blob — e2e-encrypted distributed blob storage on top of github repos. AES-256-GCM, automatic chunking, no s3 needed. stdlib only (private).
• gh-lockbox — stateless secret recovery via github actions. ephemeral gists, zero keys, auto-cleanup. rolled at n5.
• destroysaas — business portal where SMBs stop renting software and start owning it. propose a problem, recruit peers, form a cell. Wyoming LCA + DAO governance. live at destroysaas.coop.
• brane — local-first semantic knowledge graph. encode ethics, architecture constraints, and policies as machine-verifiable Datalog rules. split-brain SQLite + CozoDB.

182 public repos and counting. https://github.com/ahoward

researching AI-executable architecture — what happens when you stop writing code and start writing intent. mother is a TypeScript DSL where developers declare system boundaries, intents, and contracts, and AI agents generate the implementations. tree-sitter extracts topology, NATS routes events, a validator enforces that leaf nodes fulfill their contracts. one binary, zero deps. the central nervous system for AI-native distributed systems.

the through-line: humans declare intent, machines generate implementations, validators enforce contracts. from semantic knowledge graphs to executable architecture DSLs to cooperative ownership — the same pattern at every scale. ethics without enforcement is theater. tests are scripture, code is sand, architecture is prophecy. everything else burns.

leg’s a POS but kinda back in the mountains after 3 years, a f!ck ton of crutch work, many surgeries, many pain, much sauna, lotta PT… it continues to be slow and painful.

quietly building mountainhigh.codes - still stealth mode, still not taking clients, still building from weird places. like cabeese (plural of caboose and busen

married my jess ❤️

favorite read of late: LLMs Corrupt Your Documents When You Delegate (Laban, Schnabel, Neville). also working through michael pollan’s new one, A Mind Appears.

super digging this.

and this: