LoadingSoftware that carries weight.ONYXERA TECHLoading route
ONYXERA TECH is a product engineering studio in San Francisco. We design, build and then run web platforms, mobile apps, AI systems and the cloud underneath them — for teams who need the thing to still work in year three.
Onyxera engineering · San Francisco
Not posters on a wall. These decide what we take on, how we staff it, and the point at which we tell a client the thing they would rather not hear. Four of the six are checked by CI on every pull request.
Every engagement opens with a 14-day discovery sprint that ends in a deployed environment behind a feature flag — not a slide deck. If nothing is in front of a real user by week six, the scope is wrong and we say so on the call.
Postgres, TypeScript and a managed runtime will outlive whatever trended this morning. The novelty budget goes on the product surface, never on the infrastructure holding it up.
The engineers on the kickoff call are the engineers writing the code. We have not outsourced a line of client work since 2019, and we will decline a project outright before staffing it with people you have never met.
A closed ticket that leaves the p95 at 900ms is not done, it is closed. Teams here are measured on the number the client actually needed — activation, latency, cost per request — so we argue with requirements that will not move it, in writing, early.
Anything worth more than a day of work gets an architecture decision record: what we chose, what we rejected, and what would make us change our minds. 214 of them so far, and CI fails a pull request that skips one.
Most studios treat launch as the exit. We run managed support on the systems we build, with a paid on-call rotation and a 99.98% uptime record, because unmaintained software does not fail loudly — it fails quietly, for months.
Values that live in a deck decay. These run on every pull request, in every repo we own or maintain: a 180kb first-load budget, a 200ms p95 assertion, and a check that fails the build if an architecture decision was made without a record of it.
name: cion: [pull_request] jobs: verify: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 - run: pnpm lint && pnpm typecheck - run: pnpm test --coverage --min 80 - run: pnpm build && pnpm size-limit # 180kb first-load budget - run: onyx perf assert --p95 200ms --lcp 2.5s - run: onyx adr check --since origin/main # no silent decisionsNo pivot, no acquisition, no funding round. One rule applied for eight years straight, and the compounding that comes with it.
Onyxera starts in a sublet off Folsom Street with a single client, a logistics dashboard, and the rule we still hire against: never take on a system we would not carry the pager for.
Engineer number three joins and we inherit a React Native rebuild two vendors had already walked away from. Eleven weeks to ship. It is still in the App Store, still on our pager.
A listed fintech signs a two-year platform contract off a 14-day discovery sprint. We add a design practice the same quarter so architecture and interface stop being sequenced.
The San Francisco office opens and the company goes remote-first in the same month. Fourteen people across six countries by December, one release train, four-day cadence.
Informal retainers turn into a real product line: tiered SLAs, a paid on-call rotation, runbooks handed over on day one instead of at the end. Twenty-two systems move under contract.
The audit closes clean and opens healthcare and payments work. Every pipeline picks up signed artifacts, mandatory change review and evidence collection that runs without anyone remembering to run it.
Six engineers move full-time onto retrieval, evaluation and inference infrastructure. One rule governs the practice: nothing reaches production without an offline eval suite and a human fallback path.
The 140th product goes live — a real-time settlement console holding a 118ms p95 under peak load. Thirty-two people, eighteen countries, and the same rule we opened with.
Everyone here has shipped something and then owned it afterwards. Below is the senior half — the people you meet in week one and still work with in year three.
Founder & CEO
Product strategy · Engagement design
CTO
Distributed systems · Go · Postgres
Head of Design
Design systems · Motion · Figma tokens
Principal Engineer
TypeScript · Edge runtimes · Caching
Head of AI
Retrieval · Eval harnesses · Inference
Engineering Manager
Platform teams · Release cadence
Head of Support
SRE · SLOs · On-call rotation
Delivery Lead
Discovery · Scope control · Estimation
Senior only
Paid trial project · Answer in ten days
No whiteboard round, no algorithm trivia, no unpaid take-home. Two conversations, then a paid two-day project scoped from work we are actually doing — and the code is yours whether or not we make an offer.
We hire senior because the model only works if everyone can be trusted alone with a client and a production database. That means fewer openings, longer searches, and a real answer within ten days of the first call — including a no.
No week of laptop setup and access tickets. One command brings the whole environment up on your machine, and your first pull request is a real one, merged inside the first week. Nobody here has ever had to ask where the runbooks live.
Don’t see your role? Send us something you’ve built.
Send the messy version — the legacy system, the deadline, the thing three vendors already tried. We will tell you inside a week whether we are the right team for it.