The problem
Small stock-dependent businesses often run operations across spreadsheets, WhatsApp, memory, and disconnected tools. The owner can see numbers, but still has to work out which issue matters, whether the evidence is trustworthy, and what action is safe to take. Traditional inventory systems also expect the business to adapt to a fixed template.
What I built
I built Muana, an adaptive stock operating system that is shaped around the way each business works.
An owner describes the business, reviews the inferred people, stock, customers, money, resources, and workflows, and publishes a versioned workspace only after confirming it.
Muana then turns stock truth into a controlled action loop:
- Detect a real stock or workflow exception.
- Explain the evidence and business impact.
- Prepare the next action, such as a purchase order, production run, stock count, fulfillment step, or customer communication draft.
- Let the owner review and confirm it.
- Track whether the underlying problem was actually resolved.
Muana deliberately separates suggestions from operational truth. It can prepare configuration and drafts, but it does not invent stock, prices, payment terms, delivery outcomes, or other business facts. Important changes are tenant-scoped, versioned, previewed, and explicitly confirmed.
The workflow I am demonstrating
A customer order creates a stock shortage. Muana links the shortage to the affected order, prepares a purchase or production action using confirmed supplier or recipe evidence, waits for owner approval, records the receipt or production output through the specialist inventory workflow, allocates the replenished stock back to the original demand, and shows that the exception is resolved.
This is more than an inventory dashboard. It is a connected, evidence-backed workflow from stock truth to owner-approved action.
How I used Codex and GPT-5.6
I used Codex with GPT-5.6 throughout the implementation and verification of Muana. Claude Fable was used only as an independent reviewer to identify what had not yet been built.
Codex and GPT-5.6 helped me:
- trace the end-to-end order, shortage, purchasing or production, replenishment, fulfillment, and receivables journey;
- design tenant-isolated, idempotent workflow transitions and explicit confirmation boundaries;
- implement and repair the SvelteKit and Convex paths;
- build regression coverage for retries, provenance, operational safety, and cross-workspace isolation;
- run the type, lint, unit, Convex, production-build, and Playwright verification gates;
- turn review findings into working, tested functionality.
Primary /feedback Session ID: 019f6c24-3c7a-7aa3-97ee-21bf5923f11e
Verification
The Build Week release candidate passed:
- Svelte checks with 0 errors and 0 warnings
- backend TypeScript and ESLint
- 237 unit tests
- 278 Convex tests
- the production build
- 2 of 2 offline Playwright browser checks
External provider delivery and production webhook receipts remain separate environment proof and are not presented as completed product evidence.
What is next
My next step is to put the verified release in front of real pilot businesses and measure how often owners complete a meaningful action from trusted stock evidence.
Built With
- codex
- convex
- gpt-5.6
- playwright
- svelte-5
- sveltekit
- typescript
- vercel
- vite
- vitest