patl4588

Kai Router — Build-Mode Patch (URGENT)

Status: Ready for Nick to apply + deploy. Owner: Pat. Date: 2026-05-01.

Why

metercall.ai/start/ (Builder v2) is sending real build prompts to https://kai.metercall.ai/api/kai. The router's classifier doesn't recognize build-intent and falls through to the greeting handler — every prompt returns "Hey. What do you want to do?".

We need the classifier to detect build prompts and route them to a new build-mode handler that uses a SaaS-app-generator system prompt, runs the existing Groq cascade, and returns structured output.

patl4588

All-in-One — Flagship Module Spec (v1)

Status: v1 spec. Owner: Pat. Author: Claude (Builder). Date: 2026-05-01. For approval before implementation.

The All-in-One is the flagship module for MeterCall. It's the default product every new customer signs up for. It bundles team chat + CRM + invoicing + scheduling + projects + helpdesk + analytics into one surface, on one login, on one balance. The 90/10 keyless proxy is the differentiator — we manage the API keys for 90% of integrations; customers supply keys only for the truly sensitive 10%.

This is the proof of the entire "stop stitching" thesis. Every other module on the marketplace is a vertical branching off this base. The homepage pitch ("AI will finish the rest") lands on THIS product.

---

patl4588

Nick — marching orders (2026-05-01)

You're back online and the platform has shifted. Yesterday was QA + ops bootstrap. Today is the most important system MeterCall will build this year. Read this once. Then start.

You're not running QA on a product anymore. You're building the layer that turns autonomous code generation into trusted code generation. Without this, MeterCall ships fast slop. With it, MeterCall ships verified products at scale.

---

1. The end-state you're building

patl4588

QA Research Agent — Spec

Status: Spec for Hetzner-side daily research agent. Owner: Pat. Author: Claude. Date: 2026-05-01. Pairs with AUTO-FIX-QA-LOOP-SPEC.md v4.

This document specifies a continuous research agent that runs on Hetzner, pulls the latest cutting-edge work on autonomous QA / agent-driven testing / LLM evaluation systems, summarizes findings into a daily digest, and surfaces what we should adopt into our 6-layer system.

This is not a one-shot research task. It is a permanent intelligence loop that keeps our QA architecture at the frontier as the field evolves. While competitors copy what we shipped 3 months ago, we ship what the field is publishing today.

---

patl4588

Coaching Guide v2: Nick's Claude — Autonomous Product Intelligence

Status: v2. Operating manual for Nick's local Claude. Pairs with docs/AUTO-FIX-QA-LOOP-SPEC.md v4. Owner: Pat. Date: 2026-05-01.

This is the playbook Nick's local Claude follows to coach Nick through implementing the Build → QA → Fix → Approve → Publish → Earn loop. Nick is the operator; his Claude is the architect-coach. Pat reviews milestones; he doesn't unblock daily.

---

0. Identity

patl4588

Autonomous Product Intelligence — v4 (build spec)

Status: v4. Canonical build spec. Supersedes v1, v2, v3. Merges GPT's product-loop framing + concrete prompts/schemas + Claude's strategic + cost analysis + cross-layer logic. Owner: Pat. Forward-ready for Nick's local Claude. Date: 2026-05-01.

This document specifies the Autonomous Product Intelligence layer — the system that turns MeterCall from "fast generation" into trusted generation. It is not a QA tool. It is a core product feature equal to the builder itself.

The end-state loop:

Build → QA → Fix → Approve → Publish → Earn