The stack behind our client projects_

The frontend most of our products run on. Any React developer you hire later can pick up the codebase on day one.

One codebase for iOS and Android, shared with your web team's React skills. The Grassroots coaching app went from idea to launch in five weeks.

Builds, store releases and over-the-air updates for React Native without a native team on every release.

Where discovery turns into screens you can click through before anyone writes code.

Our default when a product needs server rendering, SEO or an installable PWA from one codebase.

Typed code end to end, so AI coding agents and new engineers break fewer things when they change it.

Fast local builds for single-page apps. Shorter feedback loops for engineers and for agents.

Design tokens in code, so screens stay consistent as the product grows and more people touch it.

Charts and maps built for your data instead of bent from a template.

Real-time 3D in the browser, no plugin. We used it to put a full aircraft livery on a microsite in three weeks.

We pick it up when your team or existing product already runs on it.

Custom shaders when off-the-shelf 3D is too heavy or too generic.

Timeline animation that stays smooth on mid-range phones.

An installable, offline-capable app without app store review. A good first release when speed matters more than native features.

Every component documented and testable on its own, so design and engineering review the same thing.

End of life since 2022. We keep AngularJS apps running and move them to React screen by screen, without a freeze on new features.

Sites built on it still work, but we now build static and content sites on Next.js.

Auth, admin and permissions come built in, so the budget goes to features that set you apart.

The language of most AI tooling, so your product backend and your AI features can share one team.

One language across frontend and backend. A good fit for real-time features and integrations.

Subscriptions, marketplace payouts and invoicing without building payments from scratch.

One typed API that web, mobile and AI agents can all query without new endpoints for every screen.

Our default database. Relational, boring in the good way, and it handles vectors when you add AI search.

Caching and queues that keep pages fast when traffic jumps.

Custom storefronts and integrations on top of Shopify's checkout.

Background tasks for Django apps: emails, reports and anything that shouldn't block a request.

In-app subscriptions and paywalls across both app stores, with revenue data from day one.

Live video and calls inside your product instead of sending users to another tool.

A headless CMS your marketing team can edit without a deploy.

Full-text and faceted search when the database alone gets slow.

Flexible documents for data that changes shape often.

Serverless Postgres that scales to zero, so an early product doesn't pay for an idle database.

Long-running jobs that survive crashes and deploys. We moved Teamdeck's queues and cron jobs onto it.

Still running in older estates. On new work and in migrations we move queues to Temporal or managed AWS services.

We maintain PHP systems and plugins, and we move legacy PHP backends to Python or Node.js when the business case is there.

Our default cloud. Your account, your bill, your data. We hand over infrastructure as code, not a console full of clicks.

Auth, push and data for mobile apps in days rather than weeks. Right for an MVP, and we know when to move off it.

The same container runs on a laptop, in CI and in production.

CI gates that every change has to pass, human or agent. Build once, promote the same artifact to production.

End-to-end tests that click through the real app, and the test-writing step in our agent harness plugins.

You hear about errors from an alert, not from a customer.

Unit tests fast enough to run on every change in the delivery loop.

Pay per request for spiky or background workloads.

Metrics, traces and logs in one place when a system gets big enough to need them.

For products with heavy, uneven traffic. Most early products don't need it, and we'll say so.

Infrastructure written in TypeScript, reviewed like any other pull request.

We work in existing Terraform estates, and we migrate them when a team wants infrastructure in its main language.

Preview deployments for every pull request. Good for Next.js products that need to move fast.

The setup we install so AI coding agents ship reviewable work: context packs, written plans, CI gates and verification. nini.bot shipped through it in 11 days.

GPT models where harder reasoning pays for the extra latency. We pick the model per task, not per vendor.

Claude Code runs the agent jobs in our open-source tools and SDLC plugins. Claude models answer fast where latency matters.

Face, emotion and eye tracking from a regular camera.

Useful glue for prototypes and custom chains. We keep it thin so it doesn't own your architecture.

Several specialised agents splitting one job, each one checkable on its own.

Answers grounded in your own documents, with sources, instead of whatever the model remembers.

Speech in, speech out, fast enough to hold a conversation. Our real-time avatar reached about 1.5 seconds of perceived latency.

A standard way to let AI assistants use your product's API, with your permissions.

We test it on long-context and multimodal tasks next to Claude and GPT.

Graph-based control for agents that need branches, retries and human approval steps.

A lighter option for tool-using agents when a full graph framework is too much.

Speech-to-speech in one model call, to cut voice latency further.