47 projects · 25 live · 0 beta · 22 lab · 0 retired
The console my fleet runs on: 52 scheduled jobs, one owner agent, Undo on everything.
proofTyped on the live page, a request shipped to production in 7 minutes and rolled back in 14 seconds.
Insurance triage in five Claude nodes; anything under 70% confidence goes to a person.
proofThe 26-case golden set went from 91.2% field and 88.5% decision accuracy to 100 and 100.
Open-source before-and-after eval harness with a noise floor and a gate that can say no.
proofFirst run rejected a token-saving filter: 10% fewer prompt tokens, but recall fell from 24 of 27 to 21 of 27.
A hosted message bus where my Claude agents and 40 Grok bots trade work, nobody relaying.
proofSchema check that blocked a deploy, answered on the bus: 17 models, 220 columns diffed, 0 mismatches.
Field-sales route planner on the App Store: offline first, a brief before every stop.
proofSolo-built and approved for the App Store; 2,925 app tests and 193 API tests passing on release day.
Home inventory for an iPhone-first user in his 60s: where is it, garage to deer lease.
proofSecond iOS app shipped end to end by the agent stack; 1.0 went live on the App Store on 2026-10-16.
AI voice receptionist for service businesses: answers, books the slot, texts the owner.
proof8 of 8 call evals pass; a live demo-line call saved the lead and texted the owner within 4 seconds.
Job-search platform for one person in HR: tailored packets, 19 job sources scanned daily.
proofFirst full scan: 3,298 postings for $0.27 of model spend, under a hard $15 a month budget guard in the code.
Claude, Gemini and GPT each draft a plan; a Claude head of council merges and runs it.
Give it a video link and it dispatches sub-agents across frontier models (Claude, Gemini, GPT) that each draft their own implementation plan; a top-tier Claude 'head of council' then synthesizes the best parts into one plan and executes and configures it on the user's machine.
A supply chain risk map you question in plain English: what if a typhoon closes Shanghai?
proofOne of 5 pinned showcase repos; the gitleaks and confidential-content scan came back with 0 findings.
Outdoor apparel store on a headless Shopify storefront with print-on-demand fulfillment.
proofOne live cutover from Bart's Tees: 121 products revendored, 6 policies renamed, voice agent 9 of 9 scenarios.
Agent-first CRM forked from open source, self-hosted behind a Cloudflare Tunnel.
proofReplaced a $500 a month hosted CRM; outbound prospect sync runs against 12,608 deduplicated contacts.
Basketball hub for my dad: live scores, calendar feed, stats, a chat that cites sources.
proofAsk-a-question chat answers from site data and web search, capped at $3 a day and $30 a month.
Outreach pipeline for rental hosts: a reel from their own photos, a letter, a human gate.
proofA 10-step Postgres state machine with an approval gate before any paid send; 76 tests.
Self-hosted NotebookLM alternative that also produces my daily digest podcast.
proofReplaced NotebookLM for the daily digest podcast; a 38-minute episode generated on the first attempt.
Publishing pipeline for Amazon KDP: production queue, keyboard review, batch approval.
proof10 pull requests from another vendor's bots were reviewed as untrusted code and merged in one weekend pass.
Payroll fraud detection: pay stubs checked for overtime, deduction and tax errors.
Analyzes pay stubs for compliance violations and identifies discrepancies in overtime, deductions, and tax calculations.
Invoice audit for small businesses: duplicates and overcharges flagged before payment.
Catches billing errors, duplicate charges, and vendor overcharges, checking invoices against contracts and historical patterns to flag discrepancies before payment.
Email drafts land in a review queue; a person approves, and an audit log records it.
Automates email workflows while maintaining human oversight.
Model releases, trending repos and tools on one dashboard, fed by Hacker News and GitHub.
Powered by real-time Hacker News and GitHub data so you never miss what's moving in AI.
Project and task manager with a command palette, Focus Mode and a live Ops Center.
Project and task manager with a command palette, Focus Mode and a live Ops Center.
Company site: services, per-prospect proposal pages, lead intake straight into the CRM.
proofLead intake behind an API key and HMAC-verified webhooks files each lead once; spec 8 of 8.
Context card for Claude Code and Codex sessions on two Macs; fetch before you edit.
proofVerified on both Macs: real sessions on each machine appear in the other's card; rule-file drift is flagged.
A proposal page per prospect, each with a chat assistant that answers as that company.
proofThe chat assistant passed 55 of 55 evals across 5 companies, backed by 69 unit tests.
Daily agent that prunes the fleet's memory store and ledgers every change it makes.
proofManual prune plus the first curator pass took my notes from 466 to 188; every change can be undone by run id.
Owner agent for Mission Control: five-minute sweeps, one healer, typed changes with Undo.
proofThree live runs shipped typed changes in 16, 8 and 7 minutes; the last was fully reverted 7 minutes later.