From One App to Seven: The Cortex Platform Story

Cortex is the platform that every Virtual Minds product sits on. It is the answer to a question we asked ourselves after shipping our second app and realizing we had built the same authentication, billing, analytics, and AI orchestration code twice — once for each product.

Where it began

In 2023, we had two apps live: Room AI and an early version of Reshot AI. Both ran on standalone Firebase projects, each with its own auth, subscription handling, push notification system, and AI integration layer. Adding a feature in one required reimplementing it in the other.

By the time we were planning the third app, the cost was clear: every new product added linear engineering load to maintain N parallel implementations of the same infrastructure.

The decision: platform or product company?

We could have stayed a portfolio of independent apps. Many companies do. The economics of shared infrastructure only pay off if you genuinely intend to keep shipping products. We made the call to invest 3 months of engineering into extracting the shared services — and from that point on, every product had to be designed to plug into the platform first, and stand alone never.

What Cortex actually is

Cortex is not one repository or one service. It is a set of design contracts that every Virtual Minds product implements:

• Identity — One auth system across all products, with cross-product session inheritance for users who use more than one app.
• Subscriptions — Unified subscription state with per-product entitlements. A user who pays for Room AI can use Garden AI features that are part of the same plan tier.
• AI orchestration — One model abstraction layer. Every product calls models through the same interface, so swapping providers happens once instead of N times.
• Analytics + events — A single event taxonomy. Every product emits the same event shapes; the internal dashboard understands them all without per-product code.
• Synapses — Inter-product control surfaces. From the internal dashboard, an operator can push an announcement, run a billing operation, or deploy a model upgrade across all products at once.

The principle that made it work

The most important architectural principle: shared services are not optional, they are the default. When a new product is proposed, the question is never "should we use the shared auth?" — it is "is there any reason we cannot use the shared auth?" And the answer must be a hard technical constraint, not preference.

This sounds obvious in retrospect. In practice it requires saying no to product teams that want to do something special. The discipline is what keeps the platform a multiplier instead of a tax.

The visible compound effect

Adding the seventh app to the platform took less engineering time than adding the third. Every new product gets identity, billing, push, analytics, support tooling, and a dashboard panel for free. The product team focuses on the actual AI feature — the part that customers pay for.

That is the platform dividend. It is also the reason we can run seven apps with a team of fewer than 10 engineers.

What Cortex becomes in 2027

The next chapter is opening Cortex to other product teams outside Virtual Minds. The same shared services that let us ship seven products with a small team should let any AI application company do the same. We are designing the API surface, billing model, and governance now. Public launch target: Q1 2027.

If you are building AI products and want to share notes on platform design, we are open to conversations.