Company
AI infrastructure engineers, obsessed with reliability
Our mission is to rebuild the operating system of AI, efficient, stable and enterprise-grade: Rebuild AI Unix. The name is the mission: Runix, from Rebuild and Unix. Runix Lab is the AI infrastructure brand of Runix AI Inc, a US-based technology company, and builds that system in five layers: storage (FS), data (Data and Pipeline), models (Models), a model gateway (Router), and the agents on top (Code and Comic), each an independent agent for its domain with its own harness.
02 · Company
What Runix is
The entity you would contract with, and what it actually operates today.
Runix AI Inc is a company incorporated in Wyoming, United States. It builds and operates infrastructure for teams putting AI into production, and sells it directly: there is no reseller between you and the people who run the service. The longer version of why it exists is Rebuild AI Unix. The original announcement from July 2026 is kept as a record of where the company started: a gateway and managed data pipelines, which became Router, Pipeline and Data. Where each product is today, and what comes next, is on the roadmap.
Seven products in five layers, at deliberately different stages. The statuses below are literal, and they are the same statuses used everywhere else on this site; none of them is aspirational.
- Runix Router: available, with accounts opened by our team. One OpenAI-compatible endpoint across many providers, with central key custody, per-key quotas and allowlists, and a log of every request with its cost. What Router does.
- Runix FS: early access, launched September 2026. A POSIX file system and multi-tier cache over the object storage a team already runs, built on the open-source Curvine project and deployed into the customer's own cloud account. What FS does.
- Runix Models: early access, by engagement. Fine-tuning and dedicated serving of small and open-weight models, evaluated against the model a team runs today before anything ships. What Models does.
- Runix Data: early access, by engagement. Training and evaluation data in six domains, with code as the focus, delivered with per-record provenance and licence and a quality report. What Data does.
- Runix Pipeline, in development with design partners. The tooling behind Runix Data: ingestion, deduplication, schema-validated extraction, PII masking that fails closed. The six Pipeline stages.
- Runix Code, in development, waitlist open. An AI coding agent for teams. Not generally available. Code and its waitlist.
- Runix Comic, in development, waitlist open. An end-to-end studio for comic dramas. Not generally available. Comic and its waitlist.
03 · The thesis
The thesis
Why one company builds all five layers, and why most teams start with one.
Why now
Every team is now an AI team, and none has an operating system
Model access, data preparation, tuning, storage and agents are bought from five vendors under five contracts, and the glue is scripts. The hard part of AI moved from the demo to the production run: keys, spend, failover, where prompts go, and what a model changed under you.
The wedge
Start with the layer that is on fire
Most teams arrive through the gateway: one endpoint, central keys, written data terms, available today. Some arrive through storage, because the bucket is starving their GPUs. The first layer earns the second; nothing requires the whole stack.
Why it compounds
Two paths, one set of interfaces
Requests go down through Router; data comes up through FS, Pipeline and Data into Models, which answer through Router again. Each layer is a product on its own, and each hand-off is an interface the team already uses, so the next layer arrives without a migration. Which layer, for which team.
04 · How we work
How we work
What we believe
The hard part of AI is not the demo: it is everything after: keeping services up when providers fail, keeping costs visible as usage grows, and turning messy data into something a model can actually use.
That is the work we love. Runix focuses on a small set of infrastructure problems and solves them properly, so the teams we serve can focus entirely on their product.
- Reliability is a feature. We design for failure first and publish no availability number we have not yet earned.
- Transparency builds trust. Clear pricing, itemized usage, honest communication.
- Long-term over short-term. We would rather keep a customer for years than win a quarter.
- Written down, not remembered. Decisions live in documents rather than in someone's head: which is also what working here is like.
Contact us
05 · In the open
What we publish openly
Three of the tools we built to run a gateway are open source, under CC0 or MIT. They work against any OpenAI-compatible endpoint, including the ones we compete with.
Awesome AI Gateway
A decision tree over 160+ AI gateways, with a reproducible cost benchmark and a compliance and supply-chain scorecard behind every entry.
CC0, no vendor money and no affiliate links. Runix Router is deliberately not listed on it — a maintainer scoring their own product beside competitors is not a score worth reading. See the list.
LLM Gateway Bench
A black-box benchmark for any gateway or relay: time-to-first-token, throughput, success rate, price multiple, plus fidelity probes for model echo, fake streaming and usage inflation.
Bring your own key and measure whoever you are evaluating — us included. Run the benchmark.
modelprobe
A dependency-free Go binary that reports, per model, whether it is up and how fast — one real but deliberately tiny call each.
Drop it in CI or a cron so an outage reaches you before it reaches your users. Get the prober.
06 · Not claimed
What we do not claim
A short list, kept deliberately. Every item on it is something we could quietly imply and have chosen not to.
No badges we have not earned
Runix does not hold, and does not assert, SOC 2, ISO 27001 or PCI status.
If your procurement process requires one of these, say so early and we will tell you plainly where we stand rather than pointing at a roadmap. What we do document.
No published uptime figure
No uptime percentage, no availability chart, no historical graph.
Publishing a number means standing behind it, and the honest version of that commitment comes with a contract, not a marketing page. What Router does when a provider fails.
No logos, no invented case studies
You will not find a customer wall or a case study on this site.
References will appear here when there are customers we are permitted to name. Until then the space stays empty rather than being filled — we would rather show you the mechanism and let you test it. How access works.
Tell us what you're building
Include your use case, expected volume, and timeline — we'll come back with a concrete plan.
Email contact@runixcloud.io