> ## Documentation Index
> Fetch the complete documentation index at: https://docs.iblueprint.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Run a prompt in chat against any model

> Open any prompt from your library in the iBlueprint chat window, swap the model on the fly, and iterate on the response without leaving the conversation.

The chat window is the fastest way to use a prompt in your library. You pick a prompt, pick a model, fill in any variables, and run it — all from the same conversation. Switching models is a one-click operation, so you can compare how the same prompt behaves on different providers without copying anything around.

## Opening a prompt in chat

There are two entry points:

1. **From the Prompt Library** — open a prompt and click **Run in chat**. A new chat tab opens with the prompt preloaded.
2. **From inside chat** — click the **library** icon in the chat composer and pick a prompt. The prompt body is inserted into the composer with variable placeholders highlighted.

## Filling in variables

If the prompt uses `{{variable_name}}` placeholders, iBlueprint shows an inputs panel above the composer. Each variable becomes a labeled field. Fill them in, then click **Run** — iBlueprint substitutes the values into the prompt body before sending it to the model.

<Tip>
  Variable values you enter are remembered per chat tab, so you can re-run the same prompt with different model choices without retyping inputs.
</Tip>

## Choosing a model

The chat window has a **model picker** in the top-right of every tab. Click it to see every model your workspace has access to, grouped by provider:

* OpenAI
* Anthropic
* Google
* Meta
* Mistral, Cohere, DeepSeek, AWS
* Any custom model you have added via [AI providers](/integrations/ai-providers), including self-hosted models behind an OpenAI-compatible endpoint

The picker shows the per-1K-token price and context window next to each model so you can pick deliberately.

### Switching mid-conversation

You can change the model at any point in a chat. Click the picker, choose a new model, and the **next** message uses the new model. Earlier messages keep the model they were originally generated with — the chat transcript shows a small badge next to each assistant message identifying which model produced it.

### Re-running against a different model

Hover over any assistant response and click **Re-run with…** to regenerate that same message against a different model. The new response appears as a sibling to the original, and you can flip between them with the arrow controls. This is the fastest way to compare how, say, Claude and GPT-4o handle the same prompt.

## Saving a chat run back to the library

If you tweaked the prompt body inline before running, click **Save changes to library** on the prompt header. iBlueprint creates a new version on the source prompt with your edits — see [Prompt collaboration](/prompts/prompt-library/collaboration) for how versions work.

If you want to keep the tweak as a separate prompt instead, click **Save as new prompt** and pick a scope.

## Running prompts programmatically

The chat experience is convenient for iteration, but the same library prompts are available from Blueprints and the API:

* **In a Blueprint** — add a `prompt` node and select **From library**. Reference the prompt by ID; updates to the library prompt flow through automatically.
* **From the API** — call `POST /v1/prompts/{id}/run` with a `model` parameter and a `variables` object. See the [API reference](/api-reference/overview) for the full schema.

## Limitations

* The chat window streams responses for any model that supports streaming. For models that do not, the full response appears at once when generation finishes.
* Vision and tool-use features depend on the selected model — the input area will hide the **Attach image** button on text-only models.
* Re-running an old message against a new model does not replay any tool calls that happened in the original run.