Introducing Cencori

Yesterday we published our thesis: AI began as something people accessed, and it is becoming something the world operates. This post is the shorter version, plus an honest account of what we have actually built so far.

Cencori is a deep technology company building the computing infrastructure global AI runs on.

Our mission is to make the infrastructure required to build and run AI accessible to everyone, everywhere.

For most people, AI is still something you access. A chatbot. An API. A model endpoint. A feature inside an application.

That is not where AI ends.

Banks will operate AI inside financial infrastructure. Hospitals will operate it across clinical systems. Factories will run it across production lines. Vehicles, robots and autonomous machines will sense their environments, decide, and act — continuously, not when a person asks.

When AI is a feature inside an application, it can be treated as a request sent to a remote model and a response returned. When AI becomes part of a factory, a hospital, a financial network or a machine, the requirements change. Latency matters differently. Reliability matters differently. The physical environment matters. The consequences of failure matter.

Where the computation happens starts to matter as much as the intelligence being computed.

More capable models do not remove that problem. They enlarge it. Better intelligence does not eliminate computing — it increases how much of the world needs it.

That is the constant underneath everything: intelligence has to run somewhere. It has to be built somewhere. And someone has to have access to the infrastructure that makes both possible.

The mission is deliberately larger than any product we have. It has to be. But a thesis with nothing running underneath it is an essay, so here is what is real right now.

The gateway. One OpenAI-compatible endpoint in front of 100+ models across every major provider. Change a base URL and your existing code works:

curl https://api.cencori.com/v1/chat/completions \
  -H "Authorization: Bearer $CENCORI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o",
    "messages": [{"role": "user", "content": "Say hello in one sentence."}]
  }'

Switch models by changing one string. Your SDK, your framework, your error handling all stay where they are.

Reliability. Automatic failover across providers. When an upstream provider degrades or goes down, requests route elsewhere and the application keeps serving. Circuit breakers keep a failing provider from taking your traffic with it.

Security on every request. PII detection and redaction, prompt injection and jailbreak detection, and content policy enforcement — running on the request before it reaches a model and on the response before it reaches your user.

Observability. Every request logged with full prompt and response, token usage, cost, latency and security results. Costs attributed by project, key and end user.

Governance. Policy enforcement with an immutable audit ledger, for organizations that have to prove what their AI did and why.

State and memory. A memory API so applications remember across conversations, and a sessions API that runs the agentic loop — including gating an autonomous tool call on human approval when policy requires it.

Beyond text. Vision, documents, image generation, speech-to-text and text-to-speech through the same gateway, the same keys, the same logs.

How you reach it. TypeScript, Python and Go SDKs, a CLI, an MCP server, and drop-in React components for the interfaces you would otherwise rebuild.

Build means giving people the computing foundation to create what comes next. Run means providing the infrastructure those systems require once they become real and increasingly embedded in the world.

Today that means the request layer: routing, security, state, observability, governance. The work ahead extends it toward the environments the thesis describes — enterprise and private infrastructure, sovereign deployments, computing closer to networks and machines, and eventually inside the machines themselves.

We are not going to pretend that exists yet. It does not. But it is the direction, and we would rather state it plainly than describe ourselves as something smaller than what we intend to build.

There are capable people everywhere. A student in Lagos. A researcher in Nairobi. An engineer in Bangalore. A robotics team in São Paulo. A founder in London. A laboratory in Tokyo.

None of us can know where the next important idea will come from. That is precisely the point. If we knew, infrastructure would not need to be broadly accessible.

So accessibility is not a marketing position for us, it is the architecture. The same foundation should hold from a first experiment to a production system serving millions, without anyone having to rebuild their computing foundation because what they built became important.

That is also why there is no monthly request cap on any plan, free included. Capping how many requests you may make is a limit on ambition, and it punishes exactly the moment a project starts working. You pay for what you use. The free tier exists to be built on, not to run out.

Plans differ by capability and support — security depth, retention, teams, governance, SSO — not by permission to make requests.

Cencori is early. The thesis describes decades of work; we are at the beginning of it. What we can say honestly today is that the request layer works, it is in production for real companies, and every week it holds more of what those companies depend on.

If you are building something with AI — an application, an agent, a model, a machine, a scientific system, or something that does not have a name yet — we would like it to run on this.

• Start building — create an account and make a request in minutes.
• Read the thesis — The Computing Infrastructure Global AI Runs On.
• Read the docs — documentation and API reference.
• Talk to us — support@cencori.com.

We do not know exactly what you will build. That is not our job. Our job is to make sure the computing is there when you are ready to build it.

The Cencori Team