fix(keenable): read result snippet from snippet, not description

The Keenable API returns both fields on every result. `description` is
frequently empty and `snippet` carries the page text, so WebSearchResult
came back with a title and a link and an empty snippet. Verified against
the live keyless endpoint: 10/10 results had `len(description) == 0` and
`len(snippet)` around 2000-2600.

Reads `snippet` first with a `description` fallback. Keenable returns
whole-page text where the other providers return a short snippet, so the
text is whitespace-collapsed (it arrives with newlines) and capped at 500
characters to keep `WebSearchResult.snippet` preview-sized; the full page
is still available through the contents step.

Adds a unit test covering the mapping with a realistic fixture (empty
`description`, populated `snippet`), the fallback, the whitespace collapse
and the cap, and link-less rows.

Author: ilya-bogin-keenable

backend/onyx/tools/tool_implementations/web_search/clients/keenable_client.py

@@ -14,6 +14,10 @@
 
 KEENABLE_DEFAULT_BASE_URL = "https://api.keenable.ai"
 KEENABLE_REQUEST_TIMEOUT_SECONDS = 30
+# Keenable returns whole-page text where the other providers return a short
+# snippet, so cap it to keep WebSearchResult.snippet preview-sized. The full
+# page is available through the contents step.
+KEENABLE_MAX_SNIPPET_CHARS = 500
 
 
 class RetryableKeenableSearchError(Exception):
@@ -113,7 +117,7 @@ def _search_with_retries(self, query: str) -> list[WebSearchResult]:
                 WebSearchResult(
                     title=(result.get("title") or "").strip(),
                     link=link,
-                    snippet=(result.get("description") or "").strip(),
+                    snippet=_extract_snippet(result),
                     author=result.get("author"),
                     published_date=None,
                 )
@@ -164,6 +168,20 @@ def test_connection(self) -> dict[str, str]:
         return {"status": "ok"}
 
 
+def _extract_snippet(result: dict[str, Any]) -> str:
+    """Pull a result's text out of the Keenable response.
+
+    Keenable returns both `snippet` and `description`. `snippet` carries the
+    page text and `description` is frequently empty, so prefer whichever has
+    content. Snippets are raw page text with newlines in them, so collapse
+    whitespace and cap the length.
+    """
+    text = " ".join(
+        str(result.get("snippet") or result.get("description") or "").split()
+    )
+    return text[:KEENABLE_MAX_SNIPPET_CHARS]
+
+
 def _build_error_message(response: requests.Response) -> str:
     return (
         f"Keenable search failed (status {response.status_code}): "

backend/tests/unit/onyx/tools/tool_implementations/websearch/test_keenable_client.py

@@ -0,0 +1,97 @@
+from typing import Any
+from unittest.mock import MagicMock
+from unittest.mock import patch
+
+from onyx.tools.tool_implementations.web_search.clients.keenable_client import (
+    KEENABLE_MAX_SNIPPET_CHARS,
+)
+from onyx.tools.tool_implementations.web_search.clients.keenable_client import (
+    KeenableClient,
+)
+
+
+def _response(results: list[dict[str, Any]]) -> MagicMock:
+    response = MagicMock()
+    response.json.return_value = {"results": results}
+    response.raise_for_status.return_value = None
+    return response
+
+
+@patch(
+    "onyx.tools.tool_implementations.web_search.clients.keenable_client.requests.post"
+)
+def test_search_reads_the_snippet_field(mock_post: MagicMock) -> None:
+    """Keenable returns both fields and `description` is frequently empty."""
+    mock_post.return_value = _response(
+        [
+            {
+                "url": "https://example.com/one",
+                "title": "One",
+                "description": "",
+                "snippet": "First page text",
+            }
+        ]
+    )
+
+    results = KeenableClient().search("test query")
+
+    assert len(results) == 1
+    assert results[0].snippet == "First page text"
+
+
+@patch(
+    "onyx.tools.tool_implementations.web_search.clients.keenable_client.requests.post"
+)
+def test_search_falls_back_to_description(mock_post: MagicMock) -> None:
+    mock_post.return_value = _response(
+        [
+            {
+                "url": "https://example.com/one",
+                "title": "One",
+                "description": "A description",
+            }
+        ]
+    )
+
+    results = KeenableClient().search("test query")
+
+    assert results[0].snippet == "A description"
+
+
+@patch(
+    "onyx.tools.tool_implementations.web_search.clients.keenable_client.requests.post"
+)
+def test_search_collapses_whitespace_and_caps_the_snippet(mock_post: MagicMock) -> None:
+    """Snippets are raw page text: newlines in them, and far longer than a snippet."""
+    mock_post.return_value = _response(
+        [
+            {
+                "url": "https://example.com/one",
+                "title": "One",
+                "description": "",
+                "snippet": "line one\n\nline two" + " padding" * 500,
+            }
+        ]
+    )
+
+    snippet = KeenableClient().search("test query")[0].snippet
+
+    assert len(snippet) == KEENABLE_MAX_SNIPPET_CHARS
+    assert "\n" not in snippet
+    assert snippet.startswith("line one line two")
+
+
+@patch(
+    "onyx.tools.tool_implementations.web_search.clients.keenable_client.requests.post"
+)
+def test_search_skips_results_without_a_link(mock_post: MagicMock) -> None:
+    mock_post.return_value = _response(
+        [
+            {"title": "No URL", "snippet": "text"},
+            {"url": "https://example.com/one", "title": "One", "snippet": "text"},
+        ]
+    )
+
+    results = KeenableClient().search("test query")
+
+    assert [result.link for result in results] == ["https://example.com/one"]