AI workflow models and variables

AI workflow models and variables in AI Flowi Workflow connect model providers and reusable values to component fields. The API hub offers managed access or a workspace’s own provider key. Supported component fields can load Global Variables by name instead of embedding a value in a workflow.

By Ai Flowi · Technical review: Janice Lee, Technical Support, Ai Flowi.

Published · Updated

Product: AI Flowi Workflow by Ai Flowi · Configuration: Model access and reusable values

Model providers in the API hub

The API hub catalogue includes OpenAI, Anthropic, Google Gemini, DeepSeek, Groq and OpenRouter. Provider routes must be configured and enabled before a model call can run. Check the catalogue shown in your workspace; a catalogue entry alone does not prove that a provider is enabled.

The Flowi AI component reads the API hub’s language-model catalogue for its Model picker. It uses the selected catalogue entry for the model call. This picker does not imply that every model or media capability from a named provider is available.

Managed access and bring your own key

Managed API hub calls use Flowi Tokens for model cost. Managed keys belong to a workspace and can restrict allowed APIs and token budgets. You do not need to place a provider key inside the Flowi AI component for managed access.

Bring your own key, or BYOK, uses a provider key saved for the workspace. The API hub stores that key encrypted and returns masked key details. The provider must match the selected model’s provider, and the saved key must be active. BYOK does not charge Flowi Tokens for provider cost; your provider account covers that cost. Workflow run fees remain separate.

Configure model access before testing a component. A valid component connection alone does not verify an external model call. See workflow usage and token history to inspect recorded charges and token balances.

Preview model-call cost before running

The API hub can estimate Flowi Token usage for a managed model call before you run it. Its estimate uses the selected rate card and an input-size preview. The configuration view shows estimated usage, a reserve amount and the balance after that reserve.

This is a preview heuristic, not the final charge for the entire workflow. The estimate does not send a model request or charge for a run. Actual managed model charges are settled through the API hub’s billing process. BYOK provider charges are separate from this managed-token preview.

Configure the Flowi AI component

  • Add the Flowi AI component to the workflow.
  • Refresh its Model picker and choose the intended API hub catalogue entry.
  • Check the selected model’s saved workspace settings in the API hub. Saved settings take precedence over the component’s Payment and BYOK entry id fields.
  • If saved settings use managed mode, the call uses managed access even when the component says byok.
  • If saved settings use byok, the call uses the saved workspace key, even when the component says managed or names another key.
  • If no settings are saved for that model, choose managed or byok in the advanced Payment field. For byok, enter the saved workspace key’s BYOK entry id, not the provider secret.
  • Before running, check the effective mode and saved key. A BYOK key must be active, belong to the workspace and match the model’s provider.
  • Test the workflow with a harmless prompt and check the model reply.

The component sends the selected model entry and requested payment mode through the API hub; saved workspace settings determine the effective mode when present. It does not store a provider key in the flow. If the model call fails, inspect the model configuration and error detail. An unavailable hub, disabled model, missing permission or unsuitable BYOK entry can prevent a call.

Existing model provider settings

Model provider settings store credentials for models used by AI Flowi Workflow. The provider service lists providers and models, validates credentials when saved, identifies configured providers, and supports enabled models and default model selection. These settings remain relevant to components that use that credential system; they are separate from a workspace BYOK entry in the API hub.

After saving provider credentials, check that the intended model is enabled and test the workflow. If the model call fails, inspect the provider configuration and the error detail.

Global Variables

Global Variables provide reusable values to supported component fields. Fields with variable loading enabled resolve a variable name to its value before the component runs. Configure the required variable before testing a component that uses it.

Keep credential values out of shared workflow text and examples. Show a variable name or placeholder when documenting a configuration, such as YOUR_MODEL_KEY. Use the field’s supported variable-loading control rather than embedding a secret in the workflow text.

Related

The LLM selector tutorial shows a documented model-routing workflow. The component catalogue helps identify its model and processing components.

Connections · First workflow · Common errors · Explore AI Flowi Workflow