Back to the overview

AI-native product engineering · Our architecture

A skeleton every product is assembled on.

Serverless on AWS, one account slice per product, building blocks on top. What it is, why we chose it, and what it costs, said plainly.

01The reason

Why a reference architecture at all.

  • No product starts from zero

    Identity, data, events, storage, delivery and observability are solved once and reused. Custom work goes to what makes your product different.

  • Cost follows use

    Scale to zero. When nobody uses it, it costs almost nothing. No reserved capacity disguised as serverless.

  • Built to change

    Modules with explicit contracts and events between them, so one part can change without rebuilding the rest.

02The modules

Ten modules, each one a concept with its purpose and its rules.

  • 01

    API

    Expose the product's interfaces.

  • 02

    Compute

    Run business logic and workflows.

  • 03

    AI

    Model access, inference, evaluation, guardrails.

  • 04

    Events

    Decouple contexts through events and queues.

  • 05

    Data

    Entities, relationships, access patterns.

  • 06

    Storage

    Documents, artifacts, large payloads.

  • 07

    Identity and access

    Authentication, authorization, tenancy.

  • 08

    Observability

    Logs, traces, metrics, cost.

  • 09

    Delivery

    Pipeline, artifacts, promotion into your accounts.

  • 10

    Frontend hosting

    Static sites, previews per change.

03Where it runs

Your product gets its own slice. Complete, isolated, portable.

  • Security

    What it holdsAudit trail and logs of the product
    Why it is separateA regulator or investor reads this one, never VM8's
  • Shared

    What it holdsPipeline, registry, DNS of the product
    Why it is separateWhat every environment needs, once
  • Dev and validation

    What it holdsPreviews per change, the validation environment
    Why it is separateTry before you ship, at near-zero cost
  • Production

    What it holdsThe live product
    Why it is separateStricter rules: regions, backups, no direct hands
  • Tenants

    What it holdsOne account per customer that needs a silo
    Why it is separateIsolation when your customers require it

Accounts are portable: a product that moves to its own organization is not rebuilt. It takes its accounts with it.

04Pros and cons

What it gives, and what it costs.

What it gives

  • Near-zero cost at rest

    Pay per request and transfer. Ideal while a product finds its market.

  • Yours, from day one

    Deployed in your name. You see every resource and can revoke our access.

  • Audit-ready

    Every product has its own evidence trail, readable by a third party without entering VM8's organization.

  • Leaves with you

    Nothing at runtime depends on VM8. Blocks are deployed software, not a service you call.

What it costs, said plainly

  • AWS, explicitly

    True serverless means AWS. Another cloud or on-premises is another architecture, not a setting. We say so up front.

  • A small fixed cost per account

    Governance tooling costs a few dollars a month per account. The price of a slice that is complete and portable.

  • Blocks are used, not owned

    You hold the right to use and configure every block version you run. The blocks themselves stay VM8's and keep evolving.

  • Event-driven thinking

    Contexts talk through events. Teams used to one big database adapt; the model built in discovery makes it natural.

05How it reaches you

From a merged change to your account, without hands on servers.

Reviewed, approvedPull requestBuild and testPipelineVersionedArtifactYour sign-offValidationYour accountProduction
  1. 01Pull request · Reviewed, approved
  2. 02Pipeline · Build and test
  3. 03Artifact · Versioned
  4. 04Validation · Your sign-off
  5. 05Production · Your account

Deployment runs with delegated, revocable access into your account. Every environment shows which version is live and what is waiting for promotion.