Main concepts in Langtail

This section explains the main concepts in Langtail. If you are new to LLMs or prompt development, start from this glossary to familiarize yourself with the terminology.

​

API keys

API Keys used to authenticate to Langtail. Can be generated here or from Deployments tab.

​

Assertion

To validate the outputs of your language models and prompts, Langtail provides three powerful assertion types: Text, Javascript and LLMs.

​

Config

A set of different prompt parameters that will be used in the test.

​

Deploy

Creates a deployment of the selected prompt version.

​

Deployment

A prompt version that is available on the Langtail API.

​

Environment

An alias to a deployment. For example, the production API URL stays the same but is served by different deployments over time.

​

Log

A single execution of a deployment. This allows you to see some metadata, the request to Langtail, the request to the provider, and the response from the provider.

​

Magic button ✨

Magic buttons are kind of an easter egg hidden all over a Langtail UI. This button will usually help you enhance your prompts, generate test cases, and do other useful stuff via LLM. Let us know how many you’ve found!

​

Member

A user in Langtail.

​

Message

A specific type of template that is only used in the playground and not deployed with the prompt version.

​

Metric

Aggregated data about activity in the prompt or project.

​

Model

A LLM model that will be used to run the prompt.

​

Output

The result of running a single test case. Either Success or Fail.

​

Parameter

A single setting for the prompt. For example, model or temperature.

​

Playground

An interface to easily edit and run a prompt. You can add templates, messages, variables, tools, and modify parameters.

​

Preview deployment

A specific environment called preview.

​

Production deployment

A specific environment called production.

​

Project

A collection of prompts. Also contains project-level API keys (shared across prompts) and an aggregated view of metrics and logs (from all prompts).

​

Prompt

A collection of parameters, templates, messages, variables, and tools. Sent as a single request to the model.

​

Prompt version

A single version of a prompt which includes everything except for the messages entered in the playground. A user can revert back to a version at any time.

​

Provider

Company or organization that develops, trains, and offers access to the underlying language model, such as OpenAI (for GPT models), Google (for LaMDA), or Anthropic (for Claude).

​

Provider key

API key used to authenticate to the provider .

​

Response

The response from the provider.

​

Staging

deployment A specific environment called staging.

​

Template

A text string that is part of a prompt. Valid types are system, assistant, user, function, and tool.

​

Test

A collection of configs, assertions and test cases. When run, each individual case is compared against every config and assertion and shows an output.

​

Test case

A user message and/or variable values which are executed with the prompt version that is being tested. When run, it returns an output.

​

Tool

A single custom function definition. Based on the user message, the provider can invoke this function to get external data or provide extra functionality.

​

Variable

Text that can be inserted into a template and changed dynamically when run.

​

Workspace

A collection of projects.

​

Workspace settings

Settings for a workspace where you can manage members and provider keys.