The platform that makes AI a multiplier, not a liability.

We’re Gantry, platform engineers who build what your team has been working around. The deployment pipeline held together with scripts, the environments nobody can reproduce, the observability you keep meaning to set up. We turn them into a platform your engineers own.

A good platform is a multiplier, not a line item.

Every engineering team says they’ll build the platform later. Later never comes. Your engineers file tickets to deploy, grep logs to debug, and treat staging as a suggestion. Self-service environments, deployment pipelines that just work, traces that pinpoint failures in seconds, these aren’t things you earn at scale. They’re what makes scale possible. And teams that go declarative now get a second tailwind: AI tools can accelerate platform delivery dramatically, but only when the engineer directing them has the systems-level understanding to ask the right questions. AI is a multiplier on expertise, not a replacement for it.

We build the platform that makes your engineers self-sufficient, so every hire ships product, not YAML.

What we do

Four multipliers, one team

LLM-Assisted Development

Deterministic guardrails make LLM-assisted development trustworthy

Most teams point an LLM at their codebase and hope. We make the machine prove itself. The wedge is cheap, deterministic guardrails almost nobody builds: custom lint and ast-grep rules that encode your conventions as enforced policy, typed contracts that fail the build when a generated client drifts, declarative environments where every machine is a pure function of one repo, preview environments per pull request, and OpenTelemetry tracing that pinpoints what the model actually changed. We run our own work this way. file-guard, our open-source Rust credential firewall, gates per-process secret reads on the laptop that built it. Every one of our machines is a pure function of a single source tree. When the guardrails are deterministic, the LLM becomes a fast contributor inside a system that catches it, not a liability you babysit.

  • Custom lint and ast-grep rules turn conventions into policy the build enforces, so an agent can’t quietly break a pattern.
  • Contract-first typed clients: the schema is the source of truth, and a drifted client fails CI, not production.
  • Declarative environments where every host is a pure function of source, so “works on my machine” stops being possible.
  • Preview environments and OpenTelemetry traces, so you see a change’s real behavior before merge, not at 2 AM.
  • We run it on ourselves: file-guard, our Rust FUSE daemon, guards credentials on the machine we build on.

Selected Work

What we've built

The products and tools we build and run. This is the bar we hold our platform work to.

Who we are

Engineers, not consultants

Gantry is a platform engineering team led by Dian, a mathematician turned systems engineer turned startup founder turned AI-native engineer.

We believe the best platform teams build systems that make every other engineer faster. The right deployment path, the right abstractions, the right observability: these aren’t overhead. They’re what let a 10-person team ship like a 50-person team. Most companies treat this as something to hire for later. We think it’s the first thing you should get right.

We stay close to the product: full-stack apps, AI pipelines, the systems that actually run on the platform. The right abstraction only becomes obvious once you’ve had to ship against it, which is why our platform work is sharper for it.

We orchestrate AI coding agents alongside human engineering judgment. Not as a gimmick, but as a disciplined workflow that compounds the speed of a small team. The agents handle the volume; we handle the decisions.

We hold ourselves to the standard we sell. Our own machines are declarative: one source tree defines every host, and our own credential-access tool runs on the laptop that ships your platform. Deterministic, declarative, reproducible is how we work before it is anything we propose.

We take the platform problems other people avoid, and we build systems that keep compounding long after we leave. We also publish the open-source tooling behind that thinking, because the best proof of platform thinking is code anyone can read.

Based in Europe. Clients everywhere.

FAQ

Frequently asked questions