docs: add Nebius Serverless endpoint RAG example

[SalikovAlex]

Related Issues

No related issue. Public Nebius searches in the deepset organization found no existing issue or PR for this example on September 18, 2026.

Proposed Changes

Clarify how OpenAIChatGenerator uses an OpenAI-compatible endpoint's URL, served model identifier and credential. Add a small RAG example using a local BM25 retriever and an existing Nebius Serverless Endpoint, with readiness, streaming, serialization and troubleshooting guidance.

Update both the next-version and current stable (3.1) component pages. This uses the existing generator; there are no provider classes, dependencies or deployment changes. The example uses Serverless Endpoints, not the Token Factory managed model API.

How did you test it?

• Executed the exact Python snippets with the real OpenAI SDK and a fake HTTP/SSE transport on Haystack main (b717d00654af0dc405e5cec35546d734c832042f, 3.2.0rc0, OpenAI 3.14.1) and stable 3.1.1 (OpenAI 3.15.0). Eight validation groups passed in each environment: request/auth/context, serialization, streaming, HTTP errors, timeout, interrupted stream, streaming limits/callback serialization, and missing credentials.
• Full Docusaurus production build passed; inspected the next and stable pages in the browser.
• Repository pre-commit checks and git diff --check passed for the changed pages.
• No live inference requests were made to a Nebius Endpoint. The local checks do not establish deployed-model output quality, ingress behavior or live SSE compatibility.

Notes for the reviewer

Keep this as a draft pending live endpoint validation. A deployment dry run passed on September 19, but two provisioning attempts returned Nebius internal errors before an endpoint was created. The endpoint must be configured for token authentication and expose the OpenAI Chat Completions API over managed HTTPS.

No release note: AGENTS.md excludes docs-only changes and .github/workflows/release_notes.yml skips the requirement for a docs: commit.

This change was developed with AI assistance. I wrote part of the code and reviewed all contributed content. The automated checks above were run. Verification of the submitting account's CLA status remains pending; any required CLA must be completed by an authorized signatory.

Checklist

• Contribution guidelines, docs guidance, agent guidance and code of conduct read.
• Conventional docs: title and commit prepared.
• Current and stable documentation updated.
• Example locally tested; pre-commit checks and docs build passed.
• Human review of all contributed content.
• Live endpoint validation.
• CLA status verified by the submitting account.

Related-issue updates, library unit tests/docstrings and a release note are not applicable to this documentation-only change.

[vercel]

@SalikovAlex is attempting to deploy a commit to the deepset Team on Vercel.

A member of the Team first needs to authorize it.

[CLAassistant]

All committers have signed the CLA.

[HaystackBot]

Hi @SalikovAlex, thanks a lot for your contribution! 🙏

We noticed that the Contributor License Agreement (CLA) check (license/cla) hasn't passed yet, so we've temporarily moved this PR to draft and paused the review assignment.

To get your PR reviewed, please sign the CLA via the link in the license/cla check below (or in the CLA bot comment). As soon as the check turns green, this PR will automatically be marked ready for review again and a reviewer will be re-assigned.

[HaystackBot]

Hi @SalikovAlex, just a friendly reminder: this PR is still in draft because the Contributor License Agreement (CLA) hasn't been signed yet. We'd love to review your contribution! Please sign the CLA via the link in the license/cla check, and this PR will automatically be marked ready for review.

[bogdankostic]

Hi @SalikovAlex, thanks for the PR. We keep provider-specific setup guidance off the core component pages, so I'm closing this one.

[SalikovAlex]

Thanks for clarifying, @bogdankostic Would a standalone Haystack cookbook example be a suitable home for the Nebius RAG walkthrough? I’d also be happy to narrow this PR to generic OpenAI-compatible endpoint guidance covering api_base_url, credentials, and served model IDs, with no provider-specific setup.