SDK overview

UserVane ships widget SDKs for in-product surveys, agent-native integrations for chat hosts, and a server adapter for Langfuse score delivery.

Widget SDKs

Package	Runtime	Role
@uservane/browser	Browser	Headless core + classic widget (UserVane.init / identify / survey)
@uservane/react	React client	Thin provider + useUserVane over the browser SDK
@uservane/react-native	React Native	Native survey card over the same browser core

Agent-native integrations

Package	Runtime	Role
@uservane/agent-core	Any JS runtime	Headless AgentController, correlation, server mint helpers
@uservane/agent-react	React client	Shared UserVaneProvider, InlineFeedback, hooks
@uservane/vercel-ai	React client	Vercel AI SDK host adapter (re-exports core + React UI)
@uservane/vercel-ai/server	Server only	mintFeedbackToken (secret key)
@uservane/openai-agents	React + server	OpenAI Agents SDK: fire-once run wrapper, client UI, Langfuse bridge
@uservane/copilotkit	React client	CopilotKit turn-resolution capture + server mint

Langfuse sync

Package	Runtime	Role
@uservane/langfuse-push	Server only	pushPendingScores Langfuse adapter

Auto-generated TypeDoc for each entry lives under API reference (TypeDoc).

Choosing a path

• In-product survey widget → browser, React, or React Native.
• Agent chat on Vercel AI SDK → @uservane/vercel-ai (+ server mint for linked sessions).
• OpenAI Agents SDK → @uservane/openai-agents (+ /server fire-once + optional /langfuse bridge).
• CopilotKit frontend → @uservane/copilotkit (backend owns Langfuse traces; mint with sessionId = threadId).
• Headless / custom host → @uservane/agent-core (+ @uservane/agent-react when you render).
• Push scores into Langfuse → @uservane/langfuse-push on your cron.

Shared rules

• Publishable key in clients; secret key only on servers (Keys).
• Show-tokens are server-issued; clients never invent them.
• Errors inside UserVane are guarded so they do not take down the host app.