docs(ollama): add streaming-with-tools example to OllamaChatGenerator reference

[albertodiazdurana]

Related Issues

• fixes Docs: surface tool-calling on the Ollama integration landing page haystack-core-integrations#3263 (follow-up; landing-page example landed in docs(ollama): add tool-calling example to integration landing page haystack-integrations#473)

Proposed Changes:

Adds a ### Streaming with Tools section to the OllamaChatGenerator reference page (docs-website/docs/pipeline-components/generators/ollamachatgenerator.mdx), between the existing ### Streaming section and the ## Usage heading.

The section includes:

• An executable example combining streaming_callback and tools on OllamaChatGenerator.
• A behavioral note: within a single generation step, streamed text tokens and tool-call deltas are mutually exclusive. When the model invokes a tool, chunk.content is empty in the streamed chunks, and the final replies[0].text is None while replies[0].tool_calls carries the reconstructed call list.

How did you test it?

Manually verified against OllamaChatGenerator + llama3.1:8b on Ollama, with two spike scripts:

1. Primary spike (directive prompt invoking get_weather): 2 chunks fired (1 carrying the tool-call delta, 1 closing). replies[0].tool_calls = [ToolCall(tool_name='get_weather', arguments={'city': 'Berlin'}, ...)]. replies[0].text is None. meta.finish_reason: stop.

2. Backfill spike (mutual-exclusivity check): six prompts spanning directive / ambiguous / arithmetic / unrelated-topic / literary. Across all six, never observed text content and tool-call deltas in the same chunk, nor text and tool_calls together in the final ChatMessage. The arithmetic prompt ("What is 2+2?") produced 99 chunks of pure text (98 text-chunks, 0 tool-chunks), confirming text streaming works as expected when the model elects not to use a tool.

The change is documentation-only and does not introduce code changes; no unit-test additions apply.

Notes for the reviewer

• The PR follows up on Docs: surface tool-calling on the Ollama integration landing page haystack-core-integrations#3263 and the companion PR docs(ollama): add tool-calling example to integration landing page haystack-integrations#473 (which adds a separate, simpler tool-calling example to the integration landing page). I authored both.
• Conventional commit title: docs(ollama): ....
• Release note: releasenotes/notes/streaming-with-tools-ollamachatgenerator-docs-8e339d62f38ebd06.yaml (single-entry enhancements, RST inline code, matches the shape of prior docs-only release notes).

Checklist

• I have read the contributors guidelines and the code of conduct.
• I have updated the related issue with new insights and changes (Docs: surface tool-calling on the Ollama integration landing page haystack-core-integrations#3263 thread updated when each PR opened).
• I have added unit tests and updated the docstrings. — N/A, docs-only PR.
• I've used one of the conventional commit types for my PR title: docs:.
• I have documented my code.
• I have added a release note file, following the contributors guidelines.
• I have run pre-commit hooks and fixed any issue. — Not run; this PR touches only docs-website/docs/.../*.mdx and releasenotes/notes/*.yaml. Happy to run pre-commit if a maintainer flags anything.

[vercel]

@albertodiazdurana 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.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
haystack-docs	Ready	Preview, Comment	May 7, 2026 8:42am

[anakin87]

Thank you!

I left some comments.

In addition, please also copy this change to 2.28 versioned docs (latest stable) in docs-website/versioned_docs/version-2.28/pipeline-components/generators/ollamachatgenerator.mdx.

[anakin87]

Since this change only affects docs, we don't need a release note. Please remove this file.

[anakin87]

Suggested change

	:::tip[What to expect when tools fire]
	When the model emits a tool call rather than free-form text, streamed chunks carry `tool_calls` deltas and `chunk.content` is empty. The final `replies[0].text` will be `None`, and `replies[0].tool_calls` holds the reconstructed call list. Plain text streaming and tool calling are mutually exclusive within a single generation step.
	:::

this is already kinda clear, so I'd remove this tip section

[albertodiazdurana]

Thanks for the review @anakin87! Addressed all three comments:

• Removed releasenotes/notes/streaming-with-tools-ollamachatgenerator-docs-8e339d62f38ebd06.yaml (docs-only change, no release note needed).
• Removed the :::tip[What to expect when tools fire] admonition at the bottom of the streaming-with-tools section, the inline comments in the example already convey the same information.
• Copied the new ### Streaming with Tools section into docs-website/versioned_docs/version-2.28/pipeline-components/generators/ollamachatgenerator.mdx (byte-identical to the v3 docs section).

Ready for another look when you have a moment.