How to use FuelR

Platform Overview

FuelR is a Prompt Management & LLM Operations platform designed for teams that need to build, test, and ship AI features with confidence. Rather than scattering prompts across notebooks or Slack threads, FuelR gives every prompt a structured home with version history, dataset-level testing, and a one-click deployment path.

The Core Workflow

Everything in FuelR follows one intentional path. Complete each step and you will have a production-grade AI feature running by the end of this guide.

LLM Settings

Before any prompt can run, FuelR needs at least one LLM provider API key. Navigate to Settings → LLM Settings in the left sidebar. This is a one-time setup — keys are encrypted at rest and only the last four characters are ever shown in the UI.

Adding a Provider

Prompts

The Prompts module is the core of FuelR. Every widget, evaluation, and log entry traces back to a prompt version. Navigate to Prompts in the sidebar and click New Prompt to get started.

Prompts come in two types:

Playground

The Playground is a sandboxed environment where you write and iterate on a prompt before it touches production traffic. It surfaces your variables as live input fields so you can test different inputs instantly.

Variables

Variables let you parameterize a prompt so the same template works across different inputs. Wrap any placeholder in double curly braces — FuelR automatically detects and renders an input field for each one.

// In your prompt text:
Summarize the following {{ document_type }} in {{ language }}:

{{ content }}

// FuelR detects three variables:
// → document_type, language, content

When you hit Run, each variable value is substituted into the prompt before it is sent to the LLM. The runtime log records the fully resolved prompt so you can always see exactly what was sent.

Versioning

Every time you click Save, FuelR creates an immutable new version of the prompt. You are always prompted for a commit message — treat these like Git commit messages.

Datasets

A single Playground run proves a prompt works on one input. A dataset proves it works across the distribution of real inputs your users send. Navigate to Datasets → New Dataset to create one.

Each dataset is a collection of test cases. A test case is a pair of:

• →input— the value injected into the prompt's primary variable
• →expected_output (optional) — ideal response, used as a reference during review

CSV Upload

The fastest way to build a dataset at scale. Prepare a .csv file with an input column and an optional expected_output column:

input,expected_output
"Charged twice, want refund","{'issue_category':'billing','severity':'high'}"
"App crashes on iOS 17","{'issue_category':'technical','severity':'critical'}"
"How do I change my email?","{'issue_category':'account','severity':'low'}"

FuelR validates the CSV before saving and surfaces any row-level errors inline — you can see exactly which rows fail and why without losing your other work.

Manual Entry

Use the row-by-row editor for small datasets (under ~20 cases) or when you want fine-grained control over each test case without preparing a spreadsheet. Both methods produce identical datasets that can be used in evaluations.

Widgets

A Widget is a self-contained, embeddable chat interface backed by one of your prompt versions. Once a prompt has been validated in the Playground and tested against a dataset, a Widget is the path to getting it in front of real users — with no custom backend work required.

Creating a Widget

Embedding the Widget

After creation, FuelR generates a <script> tag. Add it to any HTML page — the widget loads asynchronously and does not block your page rendering:

<!-- FuelR Widget -->
<script
  src="https://your-furlr-domain.com/widget.js"
  data-widget-id="YOUR_WIDGET_ID"
  defer
></script>

Runtime Logs

Every prompt execution — from the Playground, an evaluation run, or a live widget — produces a log entry. The Runtime Logs page is your real-time observatory for everything happening in your account. Navigate to Runtime Logs in the sidebar.

Logs are grouped by date for easier navigation. Use the search bar to filter by any text in the prompt or response — useful for finding a specific ticket text or isolating runs that produced a particular error string.

Log Fields

Click any row to open a side panel. Toggle between the Prompt and Response views to verify that variables were substituted correctly and that the output matches expectations.

Debugging with Logs

Runtime Logs are your primary debugging tool. Here is the workflow when something goes wrong in production:

Dashboard

The Dashboardgives you a high-level view of your account's activity. Navigate to Dashboard in the sidebar.

Use the Dashboard alongside Runtime Logs to understand both high-level spend trends and ground-level prompt quality. When cost spikes, drill into logs to find which prompt or widget is responsible.

Last updated March 2026 · FuelR Platform Documentation