CTRL by Mindmaker
Status: Current
A quieter way through AI.
CTRL is a calm AI briefing and decision partner for founders and small-team CEOs building the AI-native version of their business. Make Your Mind Up is its warm, one-question-at-a-time intake. The interface stays small while curation, memory, evidence, delivery, and AI orchestration do the heavy lifting.
Production: makeyourmindup.ai
Product: CTRL
Repository: krishanraja/mm-ctrl
Last verified: 2026-09-08 against live Supabase Brain-substrate and containment readback, plus the integrated release-candidate source tree. The public frontend remains on its earlier production baseline until PR #374 is merged and its Vercel deployment is verified.
Start here
Executable code and authoritative environment readback outrank prose. When behavior changes, update the current document in the same pull request.
Read this before touching the backend
The Supabase project is shared. CTRL runs alongside other Mindmaker surfaces in project bkyuxvschuwngtcdhsyg. Production readback on 2026-09-05 found 183 live Edge Functions. This repository contains 115 Edge Function directories excluding _shared; neither number establishes which shared-project functions CTRL owns. The database also has tables and cron jobs this repository does not own.
Two rules follow. Establish ownership and the exact live name before changing, deploying, or rolling back any function. Treat every directory under supabase/functions/ as production-significant even when this repository has no caller, because cron, external webhooks, and links in sent email may invoke it. Directory presence alone does not prove that a function is currently deployed. See architecture.
Product loop and current containment
The intended full loop is:
one-question intake
-> optional company recognition and fresh evidence
-> one-click confirmation or correction
-> consented context
-> First Lens
-> shared corroborated AI pool
-> personal ranking and briefing
-> decision weighed against evidence
-> explicit confirmation or correction
-> stronger memory
Under the 2026-09-05 trust containment, the public intake still supports text, a deterministic local result, and ordinary signup. Server voice transcription, company and profile recognition, the server-generated result, persisted handoff, result email, and creation of a new no-login briefing subscription are temporarily unavailable. The authenticated core product remains separate from those public onboarding limits.
Today, Decide, Blind Spot, Memory, Briefing, and Settings are the primary product. Context export and deeper review/build chains remain nested harnesses. The lesson-kit product is retired and /kit* redirects to /try.
Architecture at a glance
| Layer | Technology |
|---|
| Frontend | React 18, TypeScript 5.5, Vite 5.4, React Router 6 |
| UI | Tailwind CSS, Radix primitives via shadcn, Framer Motion, ctrl-ds tokens |
| Client data | TanStack Query, React Context, Supabase client |
| Backend | Supabase PostgreSQL, Auth, Storage, Edge Functions, Vault, pg_cron |
| AI | Capability-specific OpenAI, Anthropic, Gemini, and Grok paths |
| Voice | OpenAI transcription, Gemini fallback, ElevenLabs speech |
| Identity and evidence | PDL and Brandfetch for optional onboarding resolution; Perplexity, Tavily, Brave, Jina, NewsAPI.org, Exa, Artificial Analysis, RSS, GDELT, Hacker News |
| Billing and email | Stripe and Resend |
| Hosting | Vercel frontend and Supabase Cloud backend |
Measured source inventory on 2026-09-05: 115 Edge Function directories excluding _shared, 51 hook files, and 167 SQL migrations. The shared project had 183 live functions; that deployment count is not an ownership map.
There is no truthful single global “primary AI provider.” See the capability matrix in the current architecture.
Local development
Prerequisites
- Node.js
>=22 <25
- npm
- A local
.env.local with publishable frontend values
- Supabase access only for work that crosses the backend boundary
Install and run
PowerShell:
npm ci
Copy-Item .env.example .env.local
npm run dev
macOS or Linux:
npm ci
cp .env.example .env.local
npm run dev
Minimum browser variables:
VITE_SUPABASE_URLVITE_SUPABASE_ANON_KEY or VITE_SUPABASE_PUBLISHABLE_KEY
Never expose service-role, provider, payment, email, encryption, or cron secrets through VITE_* variables.
Verification
npm run docs:check
npm run standards:check
npm run typecheck
npm test -- --run
npm run build
npm run test:e2e # requires a running local or preview target
Use focused tests while iterating, then run the applicable full gates. Start npm run dev in another terminal or set E2E_BASE_URL to the authorised preview before Playwright. Most authenticated specs remain skipped until the documented auth seed helper is wired. Typecheck and changed-file lint are baseline-scoped because the repository carries disclosed historical debt. New or worsened diagnostics fail the gate.
Release safety
- Frontend releases flow through a pull request to
main; Vercel is Git-connected.
- Edge Functions deploy independently and must preserve the auth contract in
supabase/config.toml.
- Canonical production has historical migration-ledger drift, so the ledger cannot tell you what is applied. Verify by object readback instead, per release state. Never run a blanket production
supabase db push.
- Scheduled prewarm and delivery use a Vault-backed cron secret, not a database service-role setting.
- The Control Center bridge is a read-only source adapter inside the shared curation pool, not a second feed.
- A release is not verified until the source revision, deployment revision, canonical host, and real user path agree.
Follow the exact process in the replication and release guide.
Documentation governance
Current documents carry status, owner, and verification metadata. Dated roadmaps, delivery traces, old specs, and release journals are retained as history but do not compete with the current set. Run npm run docs:check to validate links, counts, decision IDs, pricing consistency, and known drift traps.
Built by Krish Raja.
