AI Overview

Build tool-using agents, streaming chat, MCP endpoints, and composable assistant UI with the AI package.

@tulip-systems/ai is the Tulip package for building application-aware AI experiences. It wraps the AI SDK with Tulip conventions for agents, reusable tools, MCP content, streaming chat, and client-side chat components.

It currently exposes:

server agents with instructions, tools, and delegation
canonical tool definitions that can resolve to agent tools or MCP tools
client-side tool renderers for chat UI
streaming chat helpers for API routes and RPC handlers
composable chat primitives for threads, messages, composers, and starter prompts
MCP content helpers for text, JSON, resources, and tables

Start here

Getting Started

Concepts

Recipes

Reference

What AI is good at

admin assistants that can route work to specialist agents
CRM, delivery, operations, and knowledge workflows with safe tool usage
streaming chat interfaces backed by server-side agents
sharing the same tool definition between agent and MCP surfaces
rendering rich tool calls in chat without importing server code into client bundles

Package entrypoints

import { createAgent, delegateToAgentTool } from "@tulip-systems/ai/agents/server";
import { ChatThread, ChatComposer } from "@tulip-systems/ai/chat/client";
import { streamAgentChatEventIterator } from "@tulip-systems/ai/chat/server";
import { text, json, table } from "@tulip-systems/ai/mcp/server";
import { defineTool, resolveAgentTool } from "@tulip-systems/ai/tools/server";
import { defineToolClient } from "@tulip-systems/ai/tools/client";

Documentation map

Getting Started: installation and first agent/chat integration
Concepts: agents, tools, chat UI, and MCP model
Recipes: concrete assistant patterns used in real admin apps
Reference: exports, public types, and gotchas