Tulip Logo IconTulip
Getting Started

Getting Started

Install the AI package and wire the smallest useful agent-backed chat flow.

This section covers the minimum moving parts for using @tulip-systems/ai in an app.

At a high level, an integration has four layers:

  1. server tools and agents
  2. a server endpoint or RPC handler that streams agent output
  3. client chat UI that sends messages and renders streamed parts
  4. optional client tool renderers for rich tool-call cards

Prerequisites

  • React 19
  • AI SDK 7
  • a model provider such as OpenRouter, OpenAI, Anthropic, or Google
  • a server boundary for agent definitions and tool execution

Mental model

The package intentionally separates server and client concerns:

  • agents/server creates tool-loop agents and delegation tools
  • tools/server defines canonical tools and resolves them to agent/MCP transports
  • chat/server streams agent responses from server code
  • chat/client renders composable chat UI
  • tools/client defines how tool calls should render in chat

On this page