AI Text

The AI Text step sends a prompt to an AI model and returns the text it writes: a summary, a reply draft, a translation, a classification. You use your own account with an AI provider, so you pay the provider directly and choose any model it offers.

Connect an AI provider

Open Connections → Add connection and choose the OpenAI-compatible tab. It works with OpenAI itself and with any server that speaks the OpenAI API: OpenRouter, Groq, Together, or a model you host.

Field	Notes
Name	optional
Base URL	OpenAI: https://api.openai.com/v1. OpenRouter: https://openrouter.ai/api/v1. The server must be reachable over HTTPS from the internet; local and private addresses are refused
Default model	the model AI Text steps use unless they pick another, for example gpt-4o-mini on OpenAI or openai/gpt-4o-mini on OpenRouter
Default image model	optional; only for AI Image steps, for example gpt-image-1
Default video model	optional; only for AI Video steps, for example google/veo-3.1 on OpenRouter
API key	your key for that provider

Alur checks the key with the provider before saving it. If it is rejected you'll see the reason and nothing is saved. The key is encrypted at rest and never shown again. Update key replaces it without breaking steps, and Disconnect removes the connection; an AI Text step still using it fails at run time.

Add the AI Text step

1. In the workflow editor, press +.
2. Pick AI Text (under AI). It starts with one of your AI connections selected; the step's form lets you choose another.
3. Fill in the fields below.

Every field accepts {{ <stepId>.<path> }} expressions, like other steps.

Field	Notes
Connection	required; the provider this step calls
Prompt	required; what you want, usually with data from earlier steps

Under Advanced (collapsed), all optional:

Field	Notes
Model	empty uses the connection's default model
System prompt	how the model should behave. Empty sends none
Temperature	0 to 2; lower is more predictable. Empty uses the provider's default
Max output tokens	caps the length of the reply. Empty uses 1024

Output and failures

On success the step's output has:

• text: the model's reply, for example {{ step1.text }}
• model: the model that answered
• usage: promptTokens, completionTokens and totalTokens, when the provider reports them

A request that is rejected (a bad key, an unknown model, no credit left) or a reply with no text fails the step with the provider's message, and the run shows it. Alur does not retry these, because each attempt would be billed. Only a rate limit (HTTP 429) and network errors are retried automatically.

Alur does not guarantee that a call is made exactly once if something fails mid-way, and every call is billed by your provider. In a For each loop, each item makes its own call, so a long list costs one request per item.