01
Recipe 01
Streaming chat
Build an assistant UI that feels alive, with messages appearing as the model responds.
useAIChat({ stream: true })
AI SDKDeveloper workbenchStreaming UI
The tiny kitfor buildingAI interfaces.
Forgeon AI SDK gives you clean primitives for chat, streaming, structured output, tools, and gateway-ready AI routes — without turning your app into a bowl of spaghetti.
pnpm add @forgeon/ai
route.ts
liveconst result = await streamText({
route: "support-chat",
input: message,
})Assistant preview
Build me a deploy checklist.
Sure. First, verify env vars, build command, runtime port, and rollback target.
Typed object
zod{
title: "Deployment Guide"
tags: ["ai", "deploy"]
ready: true
}
SDK modules
@forgeon/aiserver primitives
@forgeon/ai/reactclient hooks
@forgeon/ai/toolstyped tools
@forgeon/ai/gatewaygateway routes
AI SDK
streamText()generateObject()useAIChat()defineTool()createAgent()gatewayRoute()streamText()generateObject()useAIChat()defineTool()createAgent()gatewayRoute()streamText()generateObject()useAIChat()defineTool()createAgent()gatewayRoute()streamText()generateObject()useAIChat()defineTool()createAgent()gatewayRoute()
Field manual
The boring AI plumbing is already packed.
Most AI features need the same annoying parts: stream handling, retries, loading states, tool calls, schema validation, and provider routing. The SDK wraps those pieces into primitives you can actually enjoy using.
02
Recipe 02
Typed output
Turn unpredictable model replies into validated objects your product can actually trust.
generateObject({ schema })
03
Recipe 03
Tool actions
Let AI call product actions like search, deploy, create, summarize, or query safely.
defineTool({ input, run })
SDK Core
streamText()
useAIChat()
defineTool()
gatewayRoute()
generateObject()
createAgent()
Primitive map
Small functions. Big product behavior.
Treat the SDK like a toolbox, not a black box. Pick the primitive you need, compose it with your product flow, then let Forgeon carry the provider and gateway complexity behind the curtain.
@forgeon/ai
server primitives
@forgeon/ai/react
client hooks
@forgeon/ai/tools
typed tools
@forgeon/ai/gateway
gateway routes
Build sequence
From blank route to AI-native product flow.
01
Install the SDK
ready02
Create an AI route
ready03
Stream into UI
ready04
Add typed output
ready05
Connect tools
ready06
Ship with gateway control
readyapp/api/ai/route.ts
import { generateObject } from "@forgeon/ai"
import { z } from "zod"
export async function POST(req: Request) {
const { prompt } = await req.json()
const result = await generateObject({
route: "content-draft",
schema: z.object({
title: z.string(),
summary: z.string(),
nextSteps: z.array(z.string()),
}),
prompt,
})
return Response.json(result.object)
}Typed generation
Stop begging the model to return valid JSON.
Define the shape once, validate the response, and use predictable data in your UI. This is how AI features become product features, not fragile demos held together by hope and vibes.
schema-firstvalidated outputsafe for UIagent-ready
AI SDK
Build the AI layer like it belongs in your product.
Streaming UI, typed output, tool calls, gateway routes, and cleaner primitives for the apps you actually want to ship.
pnpm add @forgeon/ai