Source: https://katara.com/

Katara Cortex

# AI for You. Your Apps. Your Agents.

Access AI through chat, our API, or autonomous agents. Powered by a decentralized network of compute providers.

Paid in **USDC.** Settled **Onchain.**

## Chat

Use Katara Cortex directly. Pay for what you use in USDC.

[Start Chatting](https://chat.katara.com/)

## Build

Bring Katara Cortex inference into your application.

[Explore the API](https://docs.katara.com/start/api/)

## Run Agents

Agents spawn agents and pay for inference on demand with x402.

[Build with x402](https://katara.com/#agent-payments)

## Provide Compute

Host models. Serve inference. Earn USDC.

[Become a Provider](https://docs.katara.com/providers/overview/)

Earn with Katara Cortex

## Your Hardware. Your Earnings.

Put your Apple silicon Mac or supported GPU to work. Run approved AI models with Katara Cortex and get paid instantly in USDC, settled on Avalanche 🔺.

You bring the compute and set your prices. Katara Cortex connects you with demand from people, applications, and agents, and handles routing, verification, and payment.

[Become a Provider](https://docs.katara.com/providers/overview/)

Compute That Pays

### Earn in USDC.

Instant payments for the inference you serve.

-   #### Apple Silicon Macs
    
    Put your Mac’s compute to work.
    
-   #### Supported GPUs
    
    Turn GPU capacity into paid inference.
    

Available models depend on your hardware and memory.

Paid to Your Wallet

Settled on Avalanche 🔺

### Reach AI Demand

Serve inference for people, applications, and autonomous agents through one network.

### Set Your Prices

Choose what to charge for input and output tokens. Compete for requests on your terms.

### Match Your Hardware

Find approved models that fit your machine’s chip, memory, and supported runtime.

For Everyday AI

## Everyday Questions. Useful Answers.

Find a recipe for what’s in your fridge, plan your weekend, or make sense of something new. Bring everyday questions to Katara Cortex and work through them in chat.

Choose a supported model in your browser and pay as you go.

[Start Chatting](https://chat.katara.com/)

### Example Conversation

You

I have spinach, ricotta, mozzarella, and tomato sauce in the fridge, plus lasagna sheets in the pantry. Can you find me a dinner recipe?

Katara Cortex

Try spinach and ricotta lasagna. Layer cooked lasagna sheets with tomato sauce and ricotta mixed with wilted spinach. Top with mozzarella and bake until golden and bubbling.

For Developers

## One API. Real Choice.

Use the OpenAI client you already have. Point it at Katara Cortex and access supported models through the same request format.

Katara Cortex handles provider selection, routing, validation, metering, and settlement. You get the response.

-   OpenAI Compatible
-   Pay per Request in USDC
-   Onchain Settlement
-   Provider Competition
-   Managed Fallback for Supported Models

[Start Building](https://docs.katara.com/start/api/)

**POST** /v1/chat/completions

Example

model

`katara/llama-3.1-8b-instruct@1`

Response

usage

- **prompt\_tokens**: 18

- **completion\_tokens**: 52

- **total\_tokens**: 70

katara

- **cost\_usdc**: 0.000018

- **provider**: 0xfFe5…38DC

- **fallback**: false

Illustrative response from the API documentation.

For x402 Agents

## Agents That Spawn Agents.

An agent can delegate work to new agents, which can spawn agents of their own. Each can call Katara Cortex for inference and pay on the spot with x402.

Define a task, budget, and spending rules for each agent. It can buy inference as the workflow runs, paying in USDC for the work it needs to complete.

Recursive Agents.  
Inference Paid on Demand.

[View the Recursive Agents Example](https://github.com/KataraLabs/katara-recursive-agents-x402) [Read the Payment Docs](https://docs.katara.com/concepts/pricing/)

HTTP + x402

Payment Flow

1. **Request Inference** — `POST /v1/chat/completions`
2. **Receive Payment Terms** — `402 Payment Required`
3. **Authorize Payment** — `PAYMENT-SIGNATURE`
4. **Run Inference. Return a Result.** — `200 OK`

Every agent can buy inference. Paid in USDC. Settled onchain.

Inside the Network

## Know What Powers Your AI.

Katara Cortex combines approved model bundles, provider validation, and clear usage records so you can understand what serves your requests and what you’re paying for.

-   ### Verified Model Bundles
    
    Pinned model weights, tokenizers, and runtimes define what each provider must run.
    
-   ### Provider Validation
    
    Providers are tested with known prompts and checked against the bundle’s requirements.
    
-   ### Transparent Usage
    
    Request records report token usage and the cost in USDC, so you can see what you paid for.
    

[Explore Model Bundles](https://docs.katara.com/concepts/models/)

Inside a Model Bundle

### Llama 3.1 8B Instruct

`katara/llama-3.1-8b-instruct@1`

- **Bundle Version**: `@1`

A specific revision, shared by every provider serving this bundle.

- **Pinned Runtime**: `mlx-lm-0.31.3`

The runtime version is part of the published specification.

- **Weights Fingerprint**: `0xc40ca166…68f0ba0c`

The node checks the downloaded weights against the recorded hash.

Example from the published Llama 3.1 bundle. [View the Manifest](https://docs.katara.com/bundles/llama-3.1-8b-instruct.v1.json)

From the Blog

## Practical Guides for Every Path.

Explore ideas for building applications and agents, providing compute, and getting more from everyday AI.

[Explore the Blog](https://katara.com/blog)

[![](/images/blog/inference-api.webp)

Application Developers

### How to Choose an AI Inference API for Your Application

Read Article](https://katara.com/blog/choosing-an-ai-inference-api)