VoltAgent

niadraHooks() returns onPrepareModelMessages, onEnd and onHandoff for new Agent({ hooks }), and niadraTools(session) returns the kit as VoltAgent tools. VoltAgent’s own memory (conversations, working memory) stays VoltAgent’s; Niadra is the customer’s memory shared with the company’s other agents. TypeScript only.

Install

npm install @niadra/sdk @voltagent/core   # @voltagent/core 2.10 or newer, below 3, on AI SDK 6

The integration is in @niadra/sdk on npm.

The five primitives

Primitive	How the adapter wires it
Context	`onPrepareModelMessages` puts the pack (after the agent’s own notes, with `agentMemory`) as a system message after your instructions and the `turn_block` at the end of the last user message, only in what goes to the model: nothing of it lands in VoltAgent’s memory
Turns	`onPrepareModelMessages` records the user’s newest message; `onEnd` records the answer with the usage of the whole operation (prompt cache reads and writes included)
Tools	`niadraTools(session)`: VoltAgent tools with the kit’s JSON Schemas, bound to the customer
Verification	`verify` in the hooks options, recorded once before the first context
Handoff	`onHandoff` records a delegation to a subagent as a handoff between agents, when the hooks are built for a fixed conversation

The conversation comes from the operation’s context (context: { niadra: convo } on generateText or streamText), or is fixed when you pass session while building the hooks.

Minimal example

import { Agent } from "@voltagent/core";
import { openai } from "@ai-sdk/openai";
import { Niadra, handles } from "@niadra/sdk";
import { niadraHooks, niadraTools } from "@niadra/sdk/voltagent";

const niadra = new Niadra();
const support = new Agent({
  name: "support",
  instructions: "You are Acme's support agent.",
  model: openai("gpt-4.1"),
  hooks: niadraHooks(),
});

const convo = niadra.conversation({ subject: handles.appUserId(user.id), channel: "web_chat", conversation_id: chatId });
const { text } = await support.generateText("Where is my replacement?", {
  context: { niadra: convo },
  tools: niadraTools(convo),
});

Agent memory

niadraHooks({ agentMemory: true }) and niadraTools(convo, { agentMemory: true }) put the agent’s own notes before the customer’s context and add the two tools. See Agent memory.

Turn records

With the turns feature on in the space (it ships on), the adapter’s hooks with turns: true record each run as a turn record, with each tool call by the framework’s own id and each model call. The VoltAgent hooks cannot stop a tool, so wrap the tools with Niadra.tool() for a replay to answer them from the record. A turn never waits for Niadra and never repeats the conversation’s text.

Limits

VoltAgent 2.x runs on AI SDK 6; the adapter follows it.
When the model’s name does not tell the provider, pass provider in the hooks options for the usage of the agent’s turn.
Nothing here stops the agent: with Niadra slow or down, the messages go to the model as they came.
Tested against @voltagent/core 2.10.0 with a real Agent over a test model and Niadra on the emulator.

Get started

Voice

Channels

Agent frameworks

Model SDKs

Automation platforms

MCP clients

Install

The five primitives

Minimal example

Agent memory

Turn records

Limits

Next steps

Vercel AI SDK

History navigation