Inside Cencori: Building the AI Infrastructure for Developers

Discover how Cencori is building AI infrastructure for developers, the problem behind its founding, and the technology powering its vision.

The growing number of available AI models gives developers more options when building products. A team might choose one model for reasoning, another for generating images, and another for handling a different task. The challenge is managing those choices without turning the application's infrastructure into a collection of disconnected integrations.

Every additional provider can introduce more work. Developers may need to manage separate connections, handle failures, monitor usage, track costs and build controls around how their applications use AI. As more customers rely on the product, those responsibilities become harder to ignore.

The result is that teams can spend a significant amount of time maintaining the systems around their AI features instead of improving the features themselves. These are the problems that emerge when an AI experiment becomes a product people depend on. For Cencori, the goal is to make the infrastructure behind that transition easier to manage.

Founded by Bola Roy Banjo and Daniel Oreofe, Cencori is building computing infrastructure for AI, helping developers and product teams connect to AI models and manage the systems required to run them in production.

Cencori is designed to bring these operational requirements into one infrastructure layer. Rather than building and maintaining every connection independently, developers can route AI requests through Cencori and manage key parts of the process from one place.

One AI gateway, multiple models

At the centre of Cencori's platform is its AI Gateway, which sits between an application and the AI providers it uses.

Instead of connecting directly to every provider, a team can send requests through a unified interface. Cencori provides capabilities for model routing, failover, request logging, usage tracking and cost visibility. Security controls can also be configured to help teams manage the risks in what gets sent to a model.

This matters because choosing a model is only one part of operating an AI product. Teams also need to understand what happens when a provider becomes unavailable, how much requests cost, and how to monitor the behaviour of their applications over time.

A unified gateway gives developers a place to manage those concerns without having to treat every provider as an entirely separate system. For teams building AI-powered applications, that can mean less time spent maintaining integrations and more room to focus on what makes their products useful.

How a request moves through Cencori

It helps to follow one request from an application to a model and back. Each step below is a job a team would otherwise build and maintain itself.

One connection instead of many

Cencori works with the tools developers already use. Many AI apps are built on OpenAI's developer toolkit, and Cencori accepts requests in the same format. Moving an existing app over means changing two settings: where requests are sent, and which key signs them. After that, switching to a model from Anthropic, Google or any other provider Cencori supports is a matter of changing the model's name. The rest of the app stays the same.

Cencori also has its own toolkit for TypeScript, and it plugs into popular frameworks for building AI apps and agents, such as Vercel's AI SDK, LangChain and CrewAI. It isn't only for chat, either. Requests to create images, to turn speech into text and text into speech, and to prepare documents for search all go through the same gateway, with the same records and the same controls.

Whose keys pay for the call

A request can run on Cencori's own accounts with each provider, paid for from a prepaid credit balance. That gets a team started in minutes. Or a team can connect its own accounts with OpenAI, Anthropic, Google and others. Then the provider bills the team directly, the team's own usage limits apply, and it can use anything its account has access to, such as early-access or custom-trained models. Either way, Cencori still routes and records every request, along with what it cost. Teams can mix the two.

When a provider fails

AI providers have outages, slow down at busy times, and turn requests away when an account hits its limit. Cencori deals with this inside the gateway, so the app's own code doesn't have to. When a request fails for a reason that might clear up, Cencori tries again, and then sends it to a backup provider. If one provider keeps failing, Cencori stops sending it traffic until it recovers. Problems that would happen again however many times the request was sent, like a wrong key, are reported straight back so the team can fix them.

What Cencori does when a provider call fails

Problems that might clear up are retried and rerouted. Problems that will happen again are reported straight back.

When the providerTries againSwitches provider

• Is too busyIt has hit its usage limit Tries again: YesSwitches provider: Yes
• Has an error on its sideAn outage or a crash Tries again: YesSwitches provider: Yes
• Takes too long to answer Tries again: YesSwitches provider: Yes
• Rejects the keyThe account details are wrong Tries again: NoSwitches provider: No
• Rejects the requestThe request itself is wrong Tries again: NoSwitches provider: No

Teams decide what happens in a routing policy, set in Cencori's dashboard: which model to try first, which ones to fall back to, and in what order. A team might send everything to OpenAI first, then Anthropic, then Google. If OpenAI has an outage, the request goes to Anthropic instead, and the user gets an answer without ever seeing the error.

Routing can also favour the cheapest provider for a job, or whichever one is answering fastest at that moment. And when a team has several accounts with the same provider, Cencori can spread requests across them so that no single account runs into its limit.

Seeing every request

Every request through the gateway is recorded: what was sent, what came back, which model and provider answered, whether it worked, how long it took, and what it cost. When something goes wrong, a team can search these records, for example for every failed request from one key in the last hour. The dashboard turns them into charts of traffic, success rate, spending and response times, so a team can spot a slow provider or a rising bill before its customers do. Cencori says the gateway adds less than 50 milliseconds to each request.

Controls on cost and data

Spend caps stop traffic when a budget is reached, so one broken loop can't run up a large bill overnight. Alerts warn before the cap, and production, staging and internal tools can each have their own rules.

Response caching, which stays off until a team turns it on, reuses a stored answer when exactly the same request comes in again. The repeat costs nothing and comes back faster. It only applies to requests set up to give the same answer every time, so it suits jobs like sorting support tickets into categories more than open conversations.

Teams can also turn on PII protection for a project. PII is personally identifiable information, the details that point to a real person. With it on, any request containing an email address, phone number, US Social Security number or card number is blocked before it reaches a model, so customer data doesn't leave by accident.

Developers can explore the Cencori AI Gateway and its developer platform to learn more about the available tools.

Building an AI business beyond the technology

Building AI infrastructure is one part of running a business like Cencori. The other is making it possible for customers to pay for the product and access the plans that fit their needs.

Cencori uses Bachs to collect payments and manage its subscription plans, supporting the commercial side of its platform while the team focuses on building AI infrastructure for developers. From offering paid plans to collecting payments from customers, these are important parts of turning a developer tool into a business people can use and pay for.

It's easy to focus on the technology behind an AI product, but the systems that support how customers access and pay for it matter, too. For Cencori, Bachs helps support that side of the business, allowing the team to keep its attention on the larger goal: making the infrastructure needed to build and run AI accessible to more people, everywhere.

What's next?

Cencori's ambition extends beyond helping developers make requests to AI models. The company is working towards a broader infrastructure layer for how AI workloads are scheduled, executed and operated across models, providers, regions and computing resources.

That direction reflects a wider challenge in AI development. As applications become more sophisticated, teams need more than access to a model. They need infrastructure that helps them manage the systems surrounding it, from the first request to the demands of running a product for real users.

Cencori's platform brings together capabilities for developers building applications and agents, while its expanding infrastructure work points towards a future where more of the underlying complexity can be managed through a unified system.

The gateway is only part of the platform. Cencori also offers Embedded Agents, a backend for running agents for many customers at once, where each customer's data is kept apart in storage, search, billing and logs. It offers memory that lets an app remember each user across conversations, knowledge bases that ground answers in a team's own documents with citations, and a Judgment API that returns structured decisions instead of free text. In September 2026, Cencori joined NVIDIA Inception, NVIDIA's programme for startups.

The long-term ambition is to make the computing infrastructure required to build and run intelligent systems accessible to anyone, anywhere.

Cencori reports that more than 700 developers have used Cencori and that it has worked with paying organisations across AI infrastructure, model security, model hosting and compute-related use cases. These early signs of adoption sit within a much larger opportunity as developers continue to build products around AI.

The challenge ahead is helping developers build systems they can operate, monitor and sustain as their products grow. That means thinking about the infrastructure around the intelligence: how requests are routed, how failures are handled, how usage is measured, how costs are controlled and how workloads run reliably.

Cencori is building towards that future, with a goal that goes beyond providing access to models: making the infrastructure needed to build and run AI more accessible.

Explore Cencori: cencori.com

Learn more about Cencori for developers and the AI Gateway.