Introducing Seltz Answer Endpoint

We are now previewing Seltz Answer: a single endpoint that turns any natural-language prompt into a grounded, cited answer, pulled live from the web and returned as clean, streamable Markdown.

Seltz Answer sits on top of our search infrastructure, so instead of gluing together a retrieval step, a prompt, and a citation parser yourself, you send a query to our endpoint and get back an answer with the sourcing already attached. 

How it works

Under the hood, Seltz Answer runs retrieval-augmented generation (RAG) over Seltz's own search index rather than a static or cached corpus. A query comes in, the underlying search plans and executes against live results, and an LLM writes an answer grounded in what it finds, with inline citations woven directly into the response.

You can also work with the citations array in the response object, which gives you the full source list to display, audit, or re-rank on your end. Or, configure your request to include the actual document text alongside each citation and URL without a second fetch.

You can fine-tune Seltz Answers for your use case:

• Set the scope to news, people, wikipedia, or other domains that fit your question.

• Choose between a seltz-base model tier to search your exact prompt and transform the results into a natural language answer, or a seltz-pro tier to transform your prompt into a query before searching and then return a natural language answer.

Because you can choose to return streaming results as OpenAI-mimic Server-Sent Events (SSE) chunks, our Answer endpoint drops into any chat interface built for streaming completions with no adapter layer. And since the response is processed through an LLM, you can even make specific formatting requests in your prompt. This is in addition to our OpenAI-compatible chat completions endpoint.

Get grounded results from a single request

The following Python snippet makes a request against the Seltz Answer endpoint and prints out the answer along with the URL for each citation.

from seltz import Seltz
client = Seltz(api_key="YOUR_API_KEY")
response = client.answer("Who is Apple’s next CEO?", scope="news", model="seltz-base")
print(response.answer)
for citation in response.citations:
    print(citation.url)

from seltz import Seltz
client = Seltz(api_key="YOUR_API_KEY")
response = client.answer("Who is Apple’s next CEO?", scope="news", model="seltz-base")
print(response.answer)
for citation in response.citations:
    print(citation.url)

from seltz import Seltz
client = Seltz(api_key="YOUR_API_KEY")
response = client.answer("Who is Apple’s next CEO?", scope="news", model="seltz-base")
print(response.answer)
for citation in response.citations:
    print(citation.url)

There is no index to manage, no separate reranker, and no citation-stitching logic. The query goes in, and the sourced answer comes out:

Use cases for Seltz Answer

Customer-facing chat and support bots use Seltz Answer to ground responses in current information instead of a model's training data, with citations rendered inline so users can verify a claim without leaving the conversation.

News and research tools set the scope of their Answer request to news to keep grounding fresh against recently published sources, which matters most for exactly the kind of question that goes stale fast: pricing, personnel, or breaking events.

Internal knowledge and research workflows configure their Answer requests to pull full source text alongside the answer in one round trip, skipping a second fetch when the underlying document itself is what an analyst actually needs to read.

Streaming assistants enable streaming to treat Seltz Answer like any other OpenAI-style completion source, so the sourced, cited answer just becomes another token stream in an existing chat UI.

Try Seltz Answer today

Seltz Answer is available now to preview. Try it out in our playground or grab an API key at console.seltz.ai/api-keys and read the full reference at docs.seltz.ai/api-reference/answer.

Cheers, The Seltz Team