LaunchDarkly AI Configs - Skills & Cookbooks

Skills and cookbook notebooks for working with LaunchDarkly AI Configs.

Prerequisites

• A LaunchDarkly account with AI Configs enabled
• An API access token with appropriate permissions (ai-configs:write, segments:write, /*:ai-tool/*)
• Python 3.9+ installed
• Environment variable LAUNCHDARKLY_API_TOKEN set

Skills Overview

Foundation

Skill	Cookbook	Purpose
aiconfig-projects	cookbook	Create project and get SDK key

Create & Manage Configs

Skill	Cookbook	Purpose
aiconfig-tools	cookbook	Create tools for function calling
aiconfig-create	cookbook	Create AI Configs with variations
aiconfig-variations	cookbook	Manage config variations
aiconfig-update	cookbook	Update, archive, delete configs

SDK Usage

Skill	Cookbook	Purpose
aiconfig-context-basic	cookbook	Basic context patterns
aiconfig-context-advanced	cookbook	Advanced multi-context patterns
aiconfig-sdk	cookbook	Consume configs with SDK

Targeting

Skill	Cookbook	Purpose
aiconfig-segments	cookbook	Create segments for targeting
aiconfig-targeting	cookbook	Configure targeting rules

Metrics & Monitoring

Skill	Cookbook	Purpose
aiconfig-ai-metrics	cookbook	Built-in AI metrics tracking
aiconfig-online-evals	cookbook	LLM-as-a-judge evaluations
aiconfig-custom-metrics	cookbook	Custom business metrics

API Reference

Skill	Cookbook	Purpose
aiconfig-api	cookbook	API access patterns and authentication

Environment Variables

Variable	Source	Used By
LAUNCHDARKLY_API_TOKEN	LaunchDarkly UI (Authorization settings)	All API-based skills
LAUNCHDARKLY_SDK_KEY	aiconfig-projects skill output	All SDK-based skills
OPENAI_API_KEY	OpenAI	AI Metrics, Online Evals

Notes

• Agent Mode vs Completion Mode: Online evaluations (judges) only work with completion-mode configs, not agent-mode configs.
• Tools are Optional: You don't need to create tools if your AI Config doesn't use function calling.
• Enable Judges: Before using aiconfig-online-evals, enable judges in the LaunchDarkly UI.