This guide shows how to fetch Tako knowledge cards inside a Next.js API route and feed them to an LLM through the Vercel AI SDK. Copy & paste the snippets below and you’ll have a working endpoint in minutes.

​
0 · Bootstrap a Next.js project (skip if you already have one)

# Creates a Next.js 14 project (TypeScript, App Router, Tailwind pre‑configured)
npx create-next-app@latest my-tako-chat \
  --typescript --app --eslint --tailwind
cd my-tako-chat

Scaffolded files of note:

  • package.json — scripts (dev, build, start) and dependencies.
  • tsconfig.json — already set for Edge Runtime; no changes needed.
  • app/ — where your pages and API routes live.

Create a .env.local file for local runs (kept out of Git automatically):

TAKO_API_KEY=sk_tako_...
OPENAI_API_KEY=sk_openai_...

​
1 · Prerequisites

  • Node 18+
  • Next.js 14+ with the App Router (from step 0)
  • Two environment variables:
NamePurpose
TAKO_API_KEYYour key from https://trytako.com.
OPENAI_API_KEYModel provider key (swap to ANTHROPIC_API_KEY, GROQ_API_KEY, etc. as needed).

​
2 · Install packages

npm install tako-sdk ai @ai-sdk/openai
# or
pnpm add tako-sdk ai @ai-sdk/openai
# or
yarn add tako-sdk ai @ai-sdk/openai

If you’ll deploy on Vercel, push the secrets once:

npx vercel link                # one‑time project setup
vercel env add TAKO_API_KEY    # Tako
vercel env add OPENAI_API_KEY  # or ANTHROPIC_API_KEY, GROQ_API_KEY …

Why OPENAI_API_KEY? The AI SDK auto‑detects provider‑specific variables (OPENAI_API_KEY, ANTHROPIC_API_KEY, etc.), so you don’t need to reference the key in code.

​
3 · Hello‑World Edge Route

Create app/api/chat/route.ts:

import { generateText } from 'ai';
import { openai } from '@ai-sdk/openai';
import { createTakoClient } from 'tako-sdk';

export const runtime = 'edge';

export async function POST(req: Request) {
  try {
    const { question } = await req.json();
    console.log('Question:', question);

    // Fetch knowledge cards
    console.log('Fetching knowledge cards...');
    const tako = createTakoClient(process.env.TAKO_API_KEY!);
    const search = await tako.knowledgeSearch(question);
    console.log('Got ' + search.outputs.knowledge_cards.length + ' knowledge cards');

    // Extract only title and description from knowledge cards
    const knowledge_for_model = search.outputs.knowledge_cards.map(card => ({
      title: card.title,
      description: card.description
    }));

    // Ask the model, providing cards as context
    console.log('Generating response with model...');
    const { text } = await generateText({
      model: openai('gpt-4o-mini'),
      prompt: `Answer the following question using the provided knowledge cards as citations.\n\nQuestion: ${question}\n\nKnowledge cards:\n${JSON.stringify(knowledge_for_model)}`,
    });
    console.log('Model generated response:', text);

    return Response.json({
      answer: text,
      knowledge_cards: search.outputs.knowledge_cards,
    });
  } catch (error: any) {
    return Response.json(
      { error: 'Internal server error', details: error?.message || 'Unknown error' },
      { status: 500 }
    );
  }
}

Run locally:

npm run dev

Then test:

curl -X POST http://localhost:3000/api/chat \
  -H 'Content-Type: application/json' \
  -d '{"question":"How do NVIDIA and AMD revenue compare?"}' | jq

You’ll get JSON with the model’s answer and the supporting knowledge cards. Deploy with vercel --prod when you’re ready.