Cloudflare Agents

niadraAgent(this, { niadra, subject }) returns one helper per Agent instance (a Durable Object), bound to a conversation whose id is the agent’s name (the id you route to, such as a chat or a user) unless you give another. Only web APIs and the AI SDK, which the agents package already requires: it runs on Workers as it is. TypeScript only.

Install

npm install @niadra/sdk agents ai   # agents 0.24 or newer, below 1, as an optional peer dependency

The integration ships with @niadra/sdk 0.3.0, ready on main and on npm when it is published; until then the npm package is 0.1.1.

The five primitives

Primitive	How the adapter wires it
Context	`middleware` is the AI SDK middleware for `wrapLanguageModel()`, the way agents call models (`AIChatAgent.onChatMessage`, `streamText`, `generateText`): the pack after your system prompt and the `turn_block` at the end of the last user message. For a model called without the AI SDK (the Workers AI binding, a voice agent’s `onTurn()`), `prepare(messages)` returns the messages with the context in place
Turns	The middleware records the customer’s newest message and the answer with the provider’s usage; `record(text, { usage: workersAiUsage(answer, model) })` records an answer given without the AI SDK. After each answer, the queued writes go to `ctx.waitUntil()`, so the instance does not go idle before Niadra has them
Tools	`tools()`: the kit as AI SDK tools, bound to the customer
Verification	`verify` in the options, recorded once before the first context
Handoff	`end()` ends the conversation; `session.handoff()` where your agent transfers

Minimal example

import { AIChatAgent } from "@cloudflare/ai-chat";
import { convertToModelMessages, streamText, wrapLanguageModel } from "ai";
import { createWorkersAI } from "workers-ai-provider";
import { Niadra, handles } from "@niadra/sdk";
import { niadraAgent } from "@niadra/sdk/cloudflare-agents";

let niadra: Niadra | undefined; // one client per isolate

export class Support extends AIChatAgent<Env> {
  async onChatMessage() {
    niadra ??= new Niadra({ apiKey: this.env.NIADRA_API_KEY });
    const memory = niadraAgent(this, { niadra, subject: handles.appUserId(this.name) });
    const result = streamText({
      model: wrapLanguageModel({ model: createWorkersAI({ binding: this.env.AI })("@cf/openai/gpt-oss-120b"), middleware: memory.middleware }),
      system: "You are Acme's support agent.",
      messages: await convertToModelMessages(this.messages),
      tools: { ...memory.tools(), ...yourTools },
    });
    return result.toUIMessageStreamResponse();
  }
}

Agent memory

niadraAgent(this, { ..., agentMemory: true }) puts the agent’s own notes before the customer’s context and adds the two tools. See Agent memory.

Limits

The first call builds the helper; later calls on the same instance return the same one, so the conversation keeps its pinned pack and its deltas across turns.
channel: "voice" gives the voice budget and view; session takes a conversation or a task you opened yourself, in place of subject.
Nothing here stops the agent: with Niadra slow or down, the model gets the messages as they came; with no customer found, everything passes through.
Tested with the agents 0.24.0 types and, bundled with the AI SDK, running inside workerd (pnpm runtimes), with Niadra on the emulator.

Get started

Voice

Channels

Agent frameworks

Model SDKs

Automation platforms

MCP clients

Cloudflare Agents

Install

The five primitives

Minimal example

Agent memory

Limits

Next steps

Vercel AI SDK

TypeScript SDK

​Install

​The five primitives

​Minimal example

​Agent memory

​Limits

​Next steps

Vercel AI SDK

TypeScript SDK

Install

The five primitives

Minimal example

Agent memory

Limits

Next steps