Why Xava Inference comes first

An OpenRouter competitor with cashback, token-holder ownership at the centre of its vision, and a marketplace where people can sell tools for AI agents. That’s what excites me about Xava Inference.

I’m a Xava partner, which is why this is the first deployed idea I’m introducing through the Collective. It’s a business I’m already involved in, and it brings together things I expect many of our future businesses to need: access to AI models, connections to useful services, and a way to pay for the work they do.

I own a 33% share of the net value generated by Xava Inference, and that share goes to the Collective.

If we’re building a product in under four weeks and reaching first revenue in under twelve, having that foundation matters. We can spend more of our time finding customers and solving their problem.

At the time of writing, the deployed site is a private preview protected by a passcode.

An OpenRouter competitor, with cashback

OpenRouter is a single place to access AI models made by different companies. Think of it as one counter where your app can choose from a menu of AI services. You connect your app once, then choose which model to ask for help with each task.

That work is called inference: asking a trained model to write an answer, understand a document, generate an image or produce audio. Each request uses computing power and has a cost.

Xava Inference offers the same kind of convenience through one software connection, called an API, and a shared prepaid balance. The catalog covers text, images, video, audio and embeddings, which help software find information by meaning.

For a business operator, that means we can explore several ways to solve a problem through the same account. A service might need to understand a customer’s message, find the relevant information and create a useful response. A creative product might need both writing and images. Each job can use a model suited to it.

Xava handles the connection to the model, measures the usage and charges the account’s balance. The usage ledger shows what each request cost. A Playground lets us try models before building them into a product, and the OpenAI-compatible API gives supported applications a familiar connection.

The difference I want people to notice is what happens around that usage. Customers can earn cashback, and the business is designed to direct value back to its token holders.

Requests are billed at published model list prices, with no added inference markup. Negotiated savings on provider costs create the margin that funds eligible cashback. The amount depends on the model, account status and available margin; cashback can reduce the effective cost of using the service.

Card funding has its own posted price. The model catalog and billing documentation explain the current rates and terms.

A business built around token holders

The vision is a business owned by its token holders. A token is a digital asset someone can hold in a wallet. In Xava’s model, eligible holders can receive value generated by the business through token distributions and AI credits, under the rules of the token system.

What interests me is that customers, builders and supporters can also participate in the value the business creates. Someone who uses the service, brings customers or builds a useful tool has another reason to help it succeed. A community can form around both using the product and making it better.

The connection to real demand matters: people pay for useful AI work, that activity generates margin, and the margin funds the token system. The value available to share depends on the business earning it.

After inference costs and customer cashback, the direction is for 100% of the remaining margin to buy back tokens, with those tokens burned or distributed. Burning removes tokens from circulation; distributions return tokens to holders.

That is the intended direction. The preview currently allocates 10% of the remaining margin to a protocol fee and 90% to $INFX buybacks for holder distributions, burns and a permanent team lock. The full 100% model requires that allocation to change. The whitepaper describes the current arrangement.

Read the video transcript

Building with AI usually means juggling different providers. Xava Inference brings models for text, images, video and audio into one account. Like OpenRouter, with eligible cashback. A shared balance pays for each request. The vision goes further. Tools that builders can sell, and a business that shares value with token holders. I'm a Xava partner. My 33% share of net value goes to the Collective, helping build what comes next.

A marketplace for connections and tools

Composio is a useful reference for the next part: giving AI agents connections to applications and tools they can use to get things done. An agent might need to look up information, check a calendar or update a customer’s records.

Xava takes that concept into a marketplace. Builders will be able to bring their own tools, submit them for review, set a price per call and earn when people use them. Users choose which tools to enable and connect their own accounts where needed.

Imagine tools such as:

• A simple tool that lets an agent connect to your personal WhatsApp account in one click.
• A tool that gives agents access to a paid thirdweb subscription, with a small margin added to the cost.
• A research tool that searches the web and returns useful information with its sources.
• A sales tool that enriches a company record and prepares it for a CRM.
• A scheduling connection that checks availability and creates a calendar event.

These are examples of what builders could bring as the catalog develops. The preview already includes the submission, review and billing flows.

For a paid community tool, the builder sets the price they earn. Xava adds a platform fee, charges the user’s shared balance and records the builder’s earnings for payout in USDC. The platform’s margin follows the same cashback and token allocation model as inference. The tools catalog shows the current offering and pricing.

That opens up another kind of business to build with the Collective. A useful, focused tool can earn from each paid call. Its customers might be people using agents, or other businesses building those tools into their own products.

Why it belongs at the beginning

Think about a service that helps a particular kind of business handle incoming enquiries. It could use a model to understand the request, a tool to find the right information and a calendar connection to arrange the next step. We would still need to find the market, earn trust and make the workflow useful. Shared models and tools give us a faster starting point.

I’ll judge the economics by the cost of completing that customer job, including retries and human review. A cheaper response that needs fixing can cost more overall. A capable model earns its place when it helps deliver something customers will pay for.

I expect Xava Inference to support many of the ideas that follow. Some will use AI inside the product. Others may use it mainly behind the scenes, helping us research, create marketing material or run operations. The infrastructure earns its place when it helps us deliver something customers want and leaves the business with healthy economics.

The audience comes with us

The product we need today may change in six months. So may the best model for a particular job. I want us to stay able to adapt while keeping the relationships we’ve earned.

As I build in public, my audience can grow with the work. When your idea joins the portfolio, it gets a turn in that spotlight. Some of those people may become customers from the start. We still have to give them a reason to care and something worth paying for.

That is what makes this a portfolio of businesses I want to keep building: useful products, meaningful ongoing cash and people who trust us enough to see what comes next.

Xava Inference is the first project in that story, and a foundation I intend to keep building on.