kus

by 26io Techlab
Live now

Kus (Knowledge Understanding & Summarization) is an external knowledge API that makes weak or local LLMs smart — it agentically searches trusted sources, learns, and answers with citations, blending public web knowledge with your own private documents.

Open Kus →
Why Kus

Make any model answer like a frontier model.

One API that searches trusted sources and answers with citations — turning weak or local LLMs into reliable ones.

Open Kus
kus · /v1/answer
POST /v1/answer
{ "query": "What is RAG?",
  "model": "local-llama" }
// → 200 OK
{ "answer": "Retrieval-Augmented…",
  "citations": [ "arxiv.org/abs/2005.11401",
               "…" ] }
Features

What you get

Cited, verifiable answers

Agentic retrieval searches trusted sources and returns citations you can check.

Brings in your own docs

Blend private documents with public web knowledge, isolated per API key.

Real accuracy uplift

Benchmarked to make weak and local models meaningfully more accurate.

Drop-in simple

One endpoint and a key — works next to any model, in any stack.

Preview

See it in action

A real question, typed and sent to the real Kus — not scripted. From a test account, not a customer's.

See it work

A support chatbot running on a small model gets asked a policy question it doesn't actually know the answer to.

  1. 1

    Instead of guessing, the model calls Kus with the question.

  2. 2

    Kus searches trusted sources — and the business's own private documents, if any are connected — for the real answer.

  3. 3

    It returns a grounded answer with citations, which the small model can now repeat correctly and back up.

Quickstart

One request, a grounded answer

curl -X POST https://kus.26io.com/api/query \
  -H "X-API-Key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "What changed in the latest release of Next.js?",
    "return_answer": true,
    "return_chunks": true
  }'

Returns the grounded answer plus retrieval metadata — which domains it drew from, whether the response came from cache, and response time.

Full API reference
How it works

Three steps to get going

  1. 1

    Call one endpoint

    Send a question with your API key — works alongside any model you already use.

  2. 2

    Kus does the work

    It searches trusted sources, grows its knowledge base, and reranks for the best evidence.

  3. 3

    Get cited answers

    A grounded answer with citations, blending public web knowledge and your private docs.

← All 26io Techlab products
FAQ

Questions about Kus

What is Kus?

Kus is an API from 26io Techlab that makes small or local language models answer reliably. It searches trusted sources and returns answers with citations. A master router fans each query across three child retrieval systems and nineteen asynchronous source workers in parallel, then ranks results with hybrid BM25 and dense retrieval fused by reciprocal rank fusion, followed by a FlashRank cross-encoder rerank that runs on CPU. Cached questions can return in roughly 150 milliseconds with no model call.

Can Kus use an organization's own documents?

Yes. Kus can blend an organization's own private documents with public web knowledge, isolated per API key, so one customer's documents are never mixed into another's answers.

How fast are answers from Kus?

Cached questions can return in roughly 150 milliseconds with no model call. Uncached questions take longer, since Kus performs live retrieval across trusted sources and reranks the results before answering.

Where is the Kus API documentation?

Kus's public API reference is published at https://kus.26io.com/docs.