LaunchDarkly AI

LaunchDarkly AI Configs is a tool for controlling the AI models and prompts behind a live software feature. Teams store prompts and model choices outside their app code, then target specific models or prompt versions to specific users or customer segments, changing them through the LaunchDarkly interface rather than a code deploy. It is part of LaunchDarkly's AgentControl line, extending its feature-flag approach to generative AI. It supports OpenAI, Anthropic, Google Gemini, and Amazon Bedrock, plus the LangChain and LangGraph frameworks. Built-in evaluation tools score a model's cost, quality, latency, and error rate before and after a rollout, and teams can auto-escalate a request to a stronger model mid-conversation if quality drops below a threshold. Setup takes one SDK call at application startup.

This fits a company already building customer-facing AI features, like a chatbot or a summarization tool, that wants to change models or prompts fast without waiting for a full software release. A pilot on one AI feature, gating its model and prompt choice through AI Configs, is a low-risk way to test the workflow. The Developer plan is free with 5,000 AI runs a month; paid usage starts near $5 per additional 1,000 runs, the Foundation tier adds per-connection and per-user fees, and Enterprise pricing is custom, so confirm current rates since usage-based AI pricing changes often. The main limit is scope: this manages configuration and rollout, not the AI's actual output quality, so it does not replace testing prompts and models on their own merits.