β¨AI
AI Steps
Relay.app has three different AI capabilities:
Built-in actions - Common tasks like extraction, summarization, translation and transcription.
Custom Prompts - Flexible steps that can take any input and produce a desired structured output.
Mini-Agents - Semi-autonomous AI agents that can be empowered to perform certain tasks.
You can learn more about these different types of AI features in Relay.app here.
How to Connect a Model
There are two main ways to connect a model to an AI step in Relay.app.
Recommended - via Relay.app. You can use various OpenAI, Google and Anthropicmodels directly via AI credits that come with your plan in Relay.app. This means that you don't need to connect an account with an underlying model provider or purchase credits directly through that model provider. You simply select one of the first 3 options in the menu below and start using AI in your workflows. This is the simplest way for most users to interact with AI, as it means you don't need to create account in other systems, connect any API keys, or purchase credits in another product.
Advanced - via Direct Connection. If you already have an account with OpenAI, Anthropic, or another model provider, and have already purchased credits there, you can choose to call models directly via those accounts by selecting one of the Connected Models in the dialog below. Note that OpenAI platform credits are separate from ChatGPT Plus subscriptions, and you can read more about purchasing credits here.
Dealing with Errors
While AI is getting better by the day, it can still be unpredictable. In most cases, if you have a step that is resulting in an error or not producing the output you want, you'll need to adapt your prompt. We'll be adding prompt tips here in the future, but for now you can email support@relay.app for assistance.
AI Credits
Every plan in Relay.app comes with a generous amount of free AI credits. AI credits are a single pool of credits that are consumed when running steps that use AI-related features in Relay.app, including:
Prompting AI models
Audio transcription and TTS
Website scraping
The free AI credits that come with all our subscriptions are good for βreasonable use" For more details, and examples, see the tables below. If you run out of AI credits in Relay.app, you can either contact support@relay.app to purchase more, or you can switch to a Connected Model and purchase credits there.
Free monthly AI credits per plan
Adding more AI credits to your subscription
When the AI credits included in your plan aren't enough, it's very easy to add more via the billing page. Note that you need to be on a paid plan in order to be able to add additional AI credits. If you're on a Free plan, you have to upgrade to Professional or Team first.
Detailed AI credit conversion chart
Updated: 2024-09-09
AI credit usage
Overall usage
To see how many AI credits your workspace has used this period and when they refresh, look at the widget on the bottom of the left-nav menu:
Usage per step
Click on any given step in a run to see how many credits this specific step consumed:
About tokens
The definition of a token slightly differes by model provider. Additionally, input and output tokens are also differentiated from a cost/consumption perspective. We suggest reading the short FAQs from each model provider to learn more:
Unfortunately it's not trivial to predict upfront how many tokens an AI step step will consume upfront, although the article linked above will give you a better understanding.
For AI steps in Relay.app, token usage is primarily influenced by these factors:
Context that's included in the prompt. The more and larger the attached context, the more tokens will be used every time your step is executed. For example:
An Email can easily increase token usage by a few thousand
A PDF can go way beyond that, depending on the size
Whether the step is given Internet access. All text on the websites that the model needs to visit for your prompt is included as context (including for the user hidden HTML). This can be anywhere between a few hundred until 10s of thousands of tokens used per step execution, depending on how many websites the model needs to visit and how large they are
Last updated