Let's walk through a typical developer's journey building a service on top of the AI Horde:

Discovering the API

Sam is a developer who wants to add AI image generation to a web app. Instead of provisioning expensive GPU servers, Sam discovers the AI Horde's free REST API. Sam registers an account and receives an API key. With the default 25 kudos, Sam can immediately start making API calls and prototyping.

Prototyping with API calls

Sam integrates the /v2/generate/async endpoint into the app. Each image request costs kudos based on its complexity—the baseline is 10 kudos for a 512×512, 50-step image. Sam uses the dry_run payload key to estimate costs before committing, which is useful for showing users an expected wait time or cost preview without actually queuing a job.

Proxying user requests

As more users try Sam's app, the typical pattern is to have the backend act as a proxy: users interact with Sam's website, and the backend forwards their requests to the AI Horde API using Sam's own API key—users never see or need a horde account. To properly attribute usage, Sam includes the proxied_account field on /v2/generate/async requests, which tags each job with the originating user. For this to work, Sam requests the service account flag on the account by asking in the Discord server. This way, all requests benefit from Sam's kudos balance and queue priority, while the horde can still track per-user activity for abuse prevention.

Running workers to sustain the service

To build up kudos, Sam sets up two workers on cloud GPU instances during off-peak hours when rates are cheap. The workers earn kudos for every job they complete—just like any other worker on the horde. Over a weekend, the workers accumulate hundreds of thousands of kudos, giving Sam's app a large balance and high queue priority for all its users.

The give-back expectation for paid services

Sam considers offering a premium tier with faster generation. The AI Horde allows this, and the community expectation is: give back to the horde at least as much as you take out. This isn't a binding contractual obligation—it's good manners, and frankly it's just how the system naturally works. The AI Horde is built on goodwill and altruism, and the kudos system is self-limiting: if Sam's service consumes more kudos than it contributes, the account's balance drops, queue priority falls, and users lose access to certain generation parameters restricted when you have a zero-kudos accounts. In practice, running workers to replenish kudos is simply what it takes to keep the service running well. See the detailed policy on paid integrations.

Monitoring and managing costs

Sam uses the API's status endpoints to monitor worker performance, check the app's kudos balance, and track consumption. The horde tax still applies: 1 kudos per request, 1 kudos per job, plus surcharges for fast-worker-only (+20%) or blacklist (+10%) requests. Sam factors these into the app's cost estimates using dry_run.

Key things developers often miss

Proxy with a service account for multi-user apps. Have your backend proxy requests using your API key with the proxied_account field. Request the service account flag in the Discord so per-user attribution works correctly.
Workers you run serve your own requests first. If Sam's workers are online, all requests from Sam's app are automatically prioritized on those workers—a significant advantage for response times.
Concurrency is your superpower. Unlike a single local GPU, the horde can process many requests in parallel across different workers. A batch of 50 images can return much faster than generating them sequentially.
No vendor lock-in. The API is open, the code is FOSS, and you can even run a private horde instance if your needs outgrow the public one.
Kudos never expire. If Sam pauses the service for months, the accumulated kudos balance remains intact.

Frequently asked questions

Where can I get help or ask questions?

How does this all work?

Is this some sort of blockchain or cryptocurrency project?

Is this a doomed AI startup that will eventually run out of funding and shut down?

Can I request or add a feature to the Horde?

Where can I find more detailed technical information about the Horde?

Kudos

What are Kudos?

How do I earn Kudos?

How is image Kudos consumption calculated?

I don't have a powerful GPU. How can I earn Kudos?

Can I sell my Kudos?

Is Kudos a cryptocurrency?

Can you explain how a typical new user interacts with Kudos?

Can you explain how a service operator or developer would interact with Kudos?

Workers

What is a worker?

What do workers gain from participating?

Can workers see my prompts or generations?

Can I turn off my worker whenever I want?

Do workers support multiple models?

Can I prioritize myself and my friends on my own worker?

Horde

Why is it called "AI Horde"?

Can you explain the terminology used in the AI Horde?

Does the horde spy on my prompts and generations?

Why should I use the Horde instead of my local PC?

Why should I use the Horde instead of a service like Stability.ai?

Why should I use the Horde instead of a free service?

Can I run my own private horde?

Can I build paid service integrations into the AI Horde?

Not Safe for Work

Can I request NSFW images or text?

Do you censor generations?

Why are some of my images just black with white text?

Why are some of my images still censored even though I'm requesting NSFW?

Can you explain how the Anti-CSAM regex filter works?

Can you explain how the Anti-CSAM AI works?

Where can I read more about NSFW controls?