Nia

Nia by Nozomio Labs is an API for technical, up-to-date context. It lets you index docs, repos, PDFs, and datasets, then search them with AI so your agent gets real answers instead of hallucinations. The @nozomioai/nia-ai-sdk package integrates Nia with the AI SDK through tools, middleware, and direct streaming helpers.

It is designed for workflows where you want to:

Call Nia Tracer as a tool for public GitHub and documentation research
Call Nia Oracle as a tool for grounded research over indexed repositories, docs, PDFs, and datasets
Call Nia Document Agent as a tool for deep document analysis and structured extraction
Augment an existing AI SDK model with Nia-backed middleware
Stream Tracer, Oracle, or Document Agent events directly in your app

Learn more in the Nia documentation.

Setup

The Nia adapter is available in the @nozomioai/nia-ai-sdk package. You can install it with:

pnpm

npm

yarn

bun

pnpm add @nozomioai/nia-ai-sdk

npm install @nozomioai/nia-ai-sdk

yarn add @nozomioai/nia-ai-sdk

bun add @nozomioai/nia-ai-sdk

If you want to use Nia middleware with a model provider, install that provider as well. For example:

pnpm add @ai-sdk/openai

Set your Nia API key:

export NIA_API_KEY=nia_your_api_key

You can get your API key from the Nia dashboard.

Tool Usage

Nia works well as a set of AI SDK tools that you pass to generateText or streamText.

`generateText`

import { generateText } from 'ai';
import { openai } from '@ai-sdk/openai';
import { createNiaResearchTools } from '@nozomioai/nia-ai-sdk';

const tools = createNiaResearchTools({
  apiKey: process.env.NIA_API_KEY!,
  tracer: {
    defaultRequest: {
      mode: 'tracer-deep',
    },
  },
  oracle: {
    defaultRequest: {
      repositories: ['vercel/ai'],
      dataSources: ['Vercel AI SDK'],
    },
  },
  documentAgent: {
    defaultRequest: {
      sourceId: 'src_abc123',
    },
  },
});

const result = await generateText({
  model: openai('gpt-4.1'),
  prompt:
    'Research how AI SDK middleware works, then summarize the best integration pattern.',
  tools,
});

console.log(result.text);

You can pass false to any service key to disable it entirely, e.g. oracle: false if you only need Tracer. Per-service transport options (such as a separate apiKey) override the top-level ones when provided.

The SDK also exports createTracerTool(), createOracleTool(), and createDocumentAgentTool() for when you only need a single tool.

Middleware Usage

Use middleware when you want Nia to enrich the last user message before your base model runs.

`withOracleContext`

import { generateText } from 'ai';
import { openai } from '@ai-sdk/openai';
import { withOracleContext } from '@nozomioai/nia-ai-sdk';

const model = withOracleContext(openai('gpt-4.1'), {
  apiKey: process.env.NIA_API_KEY!,
  defaultRequest: {
    repositories: ['vercel/ai'],
    dataSources: ['Vercel AI SDK'],
  },
});

const result = await generateText({
  model,
  prompt: 'How should I think about AI SDK middleware for retrieval?',
});

console.log(result.text);

You can use withTracerContext() in the same way for public GitHub research and withDocumentAgentContext() when working with indexed PDFs or documents.

For more control, the lower-level createTracerMiddleware(), createOracleMiddleware(), and createDocumentAgentMiddleware() exports return raw LanguageModelMiddleware objects you can compose yourself.

Direct Streaming Helpers

If you want raw Nia job events, use the direct streaming helpers:

streamTracer()
streamOracle()
streamOracleSessionChat()
streamDocumentAgent()

import { streamTracer } from '@nozomioai/nia-ai-sdk';

const session = await streamTracer(
  {
    apiKey: process.env.NIA_API_KEY!,
  },
  {
    query: 'How does generateText stream responses?',
    repositories: ['vercel/ai'],
    mode: 'tracer-fast',
  },
);

for await (const event of session.events) {
  console.log(event.event, event.data);
}