Quickstart for AI configs
Read time: 7 minutes
Last edited: Nov 08, 2024
Overview
This topic explains how to get started with the LaunchDarkly AI configs product.
You can use AI configs to customize, test, and roll out new large language models (LLMs) and prompts within your generative AI applications. With AI configs, both the prompt and the model evaluation are specific to each end user, at runtime. You can update your prompts, specific to each end user, without redeploying your application. Follow the steps below to incorporate AI configs into your app.
Step 1, in LaunchDarkly: Create an AI config
First, create an AI config:
- Click Create and choose AI config.
- In the "Create AI config" dialog, give your AI config a human-readable Name.
- Click Create AI config.
The empty Versions tab of your new AI config displays:
Then, create a version. Every AI config has one or more versions, each of which includes your AI prompts and model configurations.
Here's how:
- In the create panel in the Versions tab, replace "Untitled version" with a version Name. You'll use this to refer to the versions when you set up targeting, below.
- Click Select a model. LaunchDarkly provides a list of common models that you can choose from.
- Enter the text of the prompt. If you'd like to customize the prompt at runtime, use
{{ example_variable }}
or{{ ldctx.example_context_attribute }}
within the prompt text. The LaunchDarkly AI SDK will substitute the correct values when you customize the AI config from within your app. - Click Save changes.
Here's an example of a completed version:
Step 2, in LaunchDarkly: Set up targeting rules
Next, set up targeting rules for your AI config. These rules determine which of your customers receive a particular version of your AI config.
To specify the AI config version to use by default when the AI config is toggled on:
- Select the Targeting tab for your AI config.
- In the "Default rule" section, click Edit.
- Configure the default rule to serve a specific version.
Here is an example of a default rule:
When you're ready to add more AI config versions, you can come back to this step and set up additional targeting rules. If you are familiar with LaunchDarkly's flag targeting, this process is very similar: with AI configs, you can target individuals or segments, or target contexts with custom rules. To learn how, read Target with AI configs.
Step 3, in your app: Install an AI SDK
First, install one of the LaunchDarkly AI SDKs in your app:
npm install @launchdarkly/server-sdk-ai
Next, import the LaunchDarkly AI client in your app and initialize a single, shared instance of it:
import { init } from '@launchdarkly/node-server-sdk';import { initAi } from '@launchdarkly/server-sdk-ai';const ldClient: LDClient = init('sdk-key-123abc');try {await ldClient.waitForInitialization({ timeout: 10 });// initialization complete} catch (error) {// timeout or SDK failed to initialize}const aiClient: LDAIClient = initAi(ldClient);
Then, set up the context. Contexts are the people or resources who will encounter generated AI in your application. The context attributes determine which version of the AI config LaunchDarkly serves to the end user, based on the targeting rules in your AI config. If you are using template variables in the prompt in your AI config's versions, the context attributes also fill in values for the template variables.
Here's how:
const context: LDContext = {kind: 'user',key: 'example-user-key',name: 'Sandy',};
Step 4, in your app: Customize the AI config, track metrics
In your code, use the modelConfig
function from the LaunchDarkly AI SDK to customize the AI config. This function returns the customized prompt and model, which you can pass directly to your AI. You can customize the prompt using both context attributes and additional variables. You should also set up a fallback value to use in case of an error.
Here's how:
const fallbackModel = {model: { modelId: 'my-default-model', 'example_variable': 'elephant' },enabled: true,};const aiConfig: LDAIConfig = aiClient.modelConfig('ai-config-key-123abc',context,fallbackModel,{ 'example_variable': 'puppy' },);// Based on the example AI config version shown in step 1,// aiConfig.config.prompt will be:// "You are a model designed to tell knock-knock jokes. Each joke should be addressed to Sandy and include at least one puppy."
Then, use one of the track[Model]
functions to record metrics from your AI model generation. LaunchDarkly provides specific functions for several common AI model families.
Here's how:
const { tracker } = aiConfig;const completion = await tracker.trackOpenAI(// Pass in the result of the OpenAI operation.// When you call the OpenAI operation, use details from aiConfigValue.// For instance, you can pass aiConfig.config.prompt// and aiConfig.config.model to your specific OpenAI operation.//// For a complete example, visit https://github.com/launchdarkly/js-core/tree/main/packages/sdk/server-ai/examples/openai.);
Step 5, in LaunchDarkly: Enable the AI config
Now that your application is set up, return to the LaunchDarkly user interface and enable your AI config. On the Targeting tab, toggle the AI config On. Then click Review and save.
When an end user opens your application, they'll get the AI config version you've defined, either in the default rule or in a custom targeting rule. Your app's AI generation will use the model and prompt from the AI config version, and the prompt will be customized for each end user.
Step 6, in LaunchDarkly: Monitor the AI config
Select the Monitoring tab for your AI config. As end users use your application, LaunchDarkly monitors the performance of your AI configs. Metrics are updated hourly:
Learn more about AI configs
The following sections provide answers to common questions about working with AI configs.
Integration with AI providers
In the AI configs product, LaunchDarkly is not handling the integration to the AI provider. The LaunchDarkly AI SDKs provide your application with model configuration details, including a customized prompt and model parameters such as temperature and tokens. It is your application's responsibility to pass this information to the AI provider.
The LaunchDarkly AI SDKs provide methods to help you track how your AI model generation is performing, and in some cases, these methods take the completion from common AI providers as a parameter. However, it is still your application's responsibility to call the AI provider. To learn more, read Tracking AI metrics.
Privacy and personally identifiable information (PII)
LaunchDarkly does not send any of the information you provide to any models, and does not use any of the information to fine tune any models.
You should follow your own organization's policies regarding if or when it may be acceptable to send end-user data either to LaunchDarkly or to an AI provider. To learn more, read AI configs and information privacy.
Availability of new models
When you create a new AI config version, you must select a model from the provided list. LaunchDarkly updates this list regularly. To request a new model, click the Give feedback option and let us know what models you'd like to have included.
What's next
We're continuing to build out the AI configs product. As the AI configs early access expands, we're committed to adding the following improvements: