Use other prompts in prompts

Prompt variables enable prompt chaining — a pattern where the output of one prompt serves as the input for another. This transforms static prompts into modular, agent-like workflows where each step in a pipeline can be independently authored, tested, and refined.

​

How it works

When you configure a variable with a Prompt source:

1. Selection — You choose another prompt from your project as the variable source.
2. Execution — At runtime, the system executes the referenced prompt first.
3. Integration — The output of the referenced prompt becomes the value of your variable, injected into the parent prompt before it is sent to the model.

​

Set up a prompt variable

​

Key capabilities

Prompt variables provide powerful orchestration features:

​

Dataset integration

When prompt variables are used with a linked dataset, they function as dynamic columns:

• Row-level execution — Each dataset row triggers its own unique prompt execution, producing a distinct result per row.
• Sequential vs. parallel — Dependent variables (where one relies on the result of another) run sequentially. Independent variables run in parallel.
• Result persistence — Outputs are stored in the dataset cells and can be used for subsequent evaluations.

​

Data flow

Child prompts can reference any column available in the parent dataset. For example, if your parent prompt has a dataset with columns userId and query, the child prompt can access both {{userId}} and {{query}}.

​

Execution behavior

​

Error handling

​

Use cases

Prompt variables are ideal for:

• Multi-step reasoning — Break complex tasks into smaller, focused prompts that execute in sequence (e.g., summarize a document, then classify the summary).
• Agent workflows — Build agent-like systems where one prompt’s output informs the next decision or action.
• Modular design — Create reusable prompt components that can be combined in different configurations.
• Output transformation — Pass the raw output of one prompt through another for formatting, translation, or filtering.

​

Best practices

• Limit nesting to 2–3 levels — Deeper hierarchies degrade performance and make debugging difficult.
• Use descriptive names — Name variables like {{user_purchase_history}} rather than {{var1}} so that the prompt chain is self-documenting.
• Test child prompts independently — Verify that each prompt in the chain works correctly on its own before connecting them.
• Watch for circular dependencies — Although Adaline detects these automatically, designing your prompt architecture to avoid them simplifies debugging.

​

Next steps