UI Guide

The Prompt Management UI is the prompt-authoring surface. Use it to write prompt templates, test them with saved scenarios, inspect rendered messages, and save versions. Promotion, targeting, and rollout still happen on the prompt’s backing managed variable.

Prompts list

Open Prompt Management from your project navigation. The list shows the prompts in the project, with their display names and slugs. Create one prompt per runtime prompt your application fetches, such as support-agent, incident-triage, or welcome-email.

Each prompt has a backing managed variable named prompt__<slug_with_underscores>. For example, a prompt with slug support-agent is fetched from application code as prompt__support_agent.

Prompt detail page

The prompt detail page has three jobs:

• author the current draft template,
• test the draft or a saved version against scenarios,
• inspect the run output before saving or promoting a version.

Template editor

The template editor is where you write the prompt text. It supports two kinds of placeholders:

Syntax	Purpose
{{customer_name}}	Runtime template input rendered from scenario variables or application data
@{support_safety_rules}@	Managed-variable composition reference expanded from Logfire configuration
@{prompt__support_style}@	Prompt composition reference expanded through another prompt’s backing variable

The editor detects template parameters and composition references. Use those detected chips to check that names match what your scenarios or managed variables provide.

Insert reference

Use Insert reference to add a managed-variable or prompt reference without typing the @{...}@ syntax by hand. Regular managed variables are inserted by their variable name. Prompts are inserted through their backing managed-variable name, such as @{prompt__support_style}@.

If you type a prompt slug without the backing prefix, Logfire can suggest the correct prompt__... name when preview or run validation fails.

Template parameters

Template parameters are the {{...}} values that must be available at render time. In the editor, those values usually come from the active scenario’s variables. In production, your application supplies the same values to logfire.template_var().get(inputs) or to its own renderer before passing the prompt to the model.

Use plain names for simple values:

customer_name = Maya Chen

Use dotted names for nested values:

customer.name = Maya Chen
customer.tier = enterprise

Dotted values are available as {{customer.name}} and {{customer.tier}}.

Scenarios panel

Scenarios are saved test cases for the prompt. A scenario contains messages and a set of variables. The default scenario starts with a system message containing:

@{prompt}@

That scenario-only alias is replaced with the rendered prompt during a run. Add user messages, prior assistant messages, and tool messages when you need to test a realistic conversation.

For example:

system: @{prompt}@
user: Customer question: {{question}}

Then define question in the scenario variables panel.

Preview and run output

Preview shows the rendered messages before they are sent to the model. It is the first place to check when output looks surprising:

• @{...}@ references should already be expanded.
• {{...}} placeholders should be replaced with scenario values.
• @{prompt}@ should be replaced inside scenario messages.
• Tool-call args and tool-return content should render string fields recursively.

If preview reports a validation error, fix the template, scenario variables, or managed-variable references before running the prompt against a model.

Versions

Saving a version freezes the prompt template text. Versions do not freeze scenario messages, model settings, tool definitions, managed-variable fragments, or rollout rules.

Use versions as the stable points you can compare, promote, or roll back to. Draft edits are not served by production traffic until you save a version and move a serving label to it.

Settings and tools

Prompt settings control the test environment used by Prompt Management. Tool definitions, model selection, and advanced gateway settings affect prompt runs in the UI. They do not become part of the prompt template your application fetches.

If your production application needs tools, configure those tools in application code as well.

Promotion

After saving a version, promote it from the backing managed variable:

1. Open the prompt’s backing variable in Managed Variables.
2. Move a label such as production, canary, or staging to the prompt version you want to serve.
3. Configure rollout or targeting rules if you want a partial rollout.

For the full workflow, see Promote and Roll Out Prompts.