Prompt Management

Prompt Management gives you a dedicated place to author the prompts that drive your LLM features, save stable versions, test them against representative inputs, and promote specific versions to production without redeploying your application.

What it is

A prompt in Logfire is a first-class piece of configuration that lives next to — but separately from — your traces. The production contract is straightforward:

• author a prompt template,
• save versions as you iterate,
• test the prompt in the editor against representative inputs, and
• promote a version for your application to consume through the Logfire SDK.

Everything else sits around that contract as testing or inspection support:

• scenarios are saved test cases,
• datasets let you sweep a scenario over many cases, and
• runs are the execution records you inspect after testing.

flowchart LR
    subgraph Prompts page
      P[Prompt]
      V[Versions]
      SC[Test scenarios]
      R[Test runs]
    end
    subgraph Variables page
      L[Labels + rollout]
    end
    SDK[Your app via SDK]

    P --> V
    P --> SC
    P --> R
    V --> L
    L --> SDK

• You author the template and, when needed, supporting test scenarios on the Prompts page.
• You save versions as you iterate — each version freezes the template text at that moment.
• You test with scenarios, datasets, and runs to see how the prompt behaves before promotion.
• You promote a version by pointing a label (for example, production) at it on the Managed Variables page for that prompt.
• Your application fetches the prompt by label through the Logfire SDK and renders the template against its runtime variables.

When to use prompts vs. the Playground

The Prompt Playground and Prompt Management solve different problems. A quick decision guide:

You want to…	Use
Explore what a one-off prompt does on a captured agent run	Prompt Playground
Tweak an existing agent run's system prompt and re-execute it	Prompt Playground
Keep a prompt that your application imports from Logfire	Prompt Management
Version a prompt, compare versions, and promote one to production	Prompt Management
Test a prompt against saved representative inputs or a dataset	Prompt Management
Give a non-engineer a stable place to iterate on production prompts	Prompt Management

The Playground is exploratory: its inputs come from a specific trace and its outputs are not persisted as first-class objects. Prompt Management is operational: prompt templates and versions become runtime configuration for your application, while scenarios, datasets, and runs are persistent testing and inspection artifacts around that runtime contract.

Where to go next

• New to the feature? Start with Concepts for the production contract (prompt, version) and the supporting testing artifacts around it.
• Writing your first template? See Templates and the full Template reference.
• Setting up saved test inputs, tool-calling rehearsal, tool definitions, or dataset runs? See Test Prompts.
• Shipping prompts from Logfire into your application? See Use Prompts in Your Application.