Knowledge bases

OverviewCopy Anchor Link

Knowledge bases let you index your own documents so the Unii API can ground chat completions in your content. Instead of relying on the model's training data, responses are based on the documents you provide — reducing hallucinations and improving accuracy.

How knowledge bases workCopy Anchor Link

1. Create a knowledge base and give it a name.
2. Upload documents — text files, markdown, PDFs, or structured data.
3. Query — reference the knowledge base in your chat completion requests. Unii handles chunking, embedding, and retrieval automatically.

Creating a knowledge baseCopy Anchor Link

import { UniiClient } from "@unii/sdk";const client = new UniiClient({apiKey: process.env.UNII_API_KEY,});const kb = await client.knowledgeBases.create({name: "Product Documentation",description: "Technical docs for our product",});console.log(kb.id); // "kb_abc123"

Adding documentsCopy Anchor Link

After creating a knowledge base, add documents:

const document = await client.knowledgeBases.documents.create(kb.id, {filename: "getting-started.md",content: fs.readFileSync("./docs/getting-started.md", "utf-8"),});console.log(document.status); // "indexing" → "ready"

Using a knowledge base in chatCopy Anchor Link

Reference the knowledge base when creating a chat completion:

const response = await client.createChatCompletion({model: "unii-latest",knowledgeBaseId: "kb_abc123",messages: [  { role: "user", content: "How do I reset my password?" },],});console.log(response.choices[0].message.content);

The response is grounded in the documents you indexed, not the model's general training data.

Supported formatsCopy Anchor Link

Knowledge bases accept the following document formats:

• Markdown (.md, .mdx)
• Plain text (.txt)
• PDF (.pdf)
• HTML (.html)
• JSON (.json)

Next stepsCopy Anchor Link

• Try chat completions to send grounded requests.
• Enable streaming for real-time responses.
• Read the API Reference for the full knowledge base API.