Skip to content

RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.

fix(agent/tools): read Keenable result content from `snippet` (#18341)

Ilya Bogin9c81d9be8b4d
diff --git a/agent/tools/keenable.py b/agent/tools/keenable.py
index 3797336d0e..f2d7ed6f9a 100644
--- a/agent/tools/keenable.py
+++ b/agent/tools/keenable.py
@@ -56,6 +56,17 @@ def _request(method: str, public_path: str, keyed_path: str, api_key: str, *, pa
return resp.json()
+def _result_content(result: dict) -> 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. The chunks built from this feed retrieval, so the text is kept
+ whole; only the newlines it arrives with are normalized.
+ """
+ return " ".join(str(result.get("snippet") or result.get("description") or "").split())
+
+
class KeenableSearchParam(ToolParamBase):
"""
Define the Keenable search component parameters.
@@ -149,7 +160,7 @@ class KeenableSearch(ToolBase, ABC):
results,
get_title=lambda r: r.get("title"),
get_url=lambda r: r.get("url"),
- get_content=lambda r: r.get("description"),
+ get_content=_result_content,
)
self.set_output("json", results)
logging.info(f"KeenableSearch: returned {len(results)} results")
diff --git a/test/unit_test/agent/tools/test_keenable_unit.py b/test/unit_test/agent/tools/test_keenable_unit.py
new file mode 100644
index 0000000000..d3a915c2b0
--- /dev/null
+++ b/test/unit_test/agent/tools/test_keenable_unit.py
@@ -0,0 +1,150 @@
+#
+# Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+"""Result-mapping regression tests for the Keenable agent tool.
+
+The Keenable API returns both ``description`` and ``snippet`` on every result.
+``description`` is frequently empty and ``snippet`` carries the page text, so
+reading ``description`` handed retrieval a title and a URL with no content at
+all. These tests apply the ``get_content`` getter the way ``_retrieve_chunks``
+does, since that getter is where the bug lived.
+
+``agent.tools.keenable`` is loaded in isolation (its package ``__init__`` would
+auto-discover every tool and pull in the full agent framework), with the agent
+base classes stubbed so only the real mapping runs.
+"""
+
+import importlib.util
+import sys
+import types
+from pathlib import Path
+from types import SimpleNamespace
+
+_REPO_ROOT = Path(__file__).resolve().parents[4]
+
+
+def _load_keenable_module():
+ base = types.ModuleType("agent.tools.base")
+
+ class _ToolParamBase:
+ def __init__(self):
+ pass
+
+ class _ToolBase:
+ def __init__(self, *a, **k):
+ pass
+
+ base.ToolParamBase = _ToolParamBase
+ base.ToolBase = _ToolBase
+ base.ToolMeta = dict
+ for pkg in ("agent", "agent.tools"):
+ sys.modules.setdefault(pkg, types.ModuleType(pkg))
+ sys.modules["agent.tools.base"] = base
+
+ # Neutralize the @timeout decorator so _invoke is a plain method.
+ conn_utils = types.ModuleType("common.connection_utils")
+ conn_utils.timeout = lambda *a, **k: lambda f: f
+ sys.modules["common.connection_utils"] = conn_utils
+
+ spec = importlib.util.spec_from_file_location("keenable_uut", _REPO_ROOT / "agent" / "tools" / "keenable.py")
+ mod = importlib.util.module_from_spec(spec)
+ spec.loader.exec_module(mod)
+ return mod
+
+
+_keenable_mod = _load_keenable_module()
+KeenableSearch = _keenable_mod.KeenableSearch
+
+
+def _build_tool():
+ cpn = KeenableSearch.__new__(KeenableSearch)
+ cpn._canvas = SimpleNamespace()
+ cpn._param = SimpleNamespace(mode="pro", top_n=10, api_key="", max_retries=0, delay_after_error=0)
+ cpn.check_if_canceled = lambda *a, **k: False
+
+ captured = {}
+ out = {}
+
+ def fake_retrieve(res_list, get_title=None, get_url=None, get_content=None, **_kw):
+ # The real _retrieve_chunks applies these getters to every result;
+ # replicate that so the tests actually exercise them.
+ items = list(res_list)
+ captured["rendered"] = [{"title": get_title(r), "url": get_url(r), "content": get_content(r)} for r in items]
+ out["formalized_content"] = "FC"
+
+ cpn._retrieve_chunks = fake_retrieve
+ cpn.set_output = lambda key, value: out.__setitem__(key, value)
+ cpn.output = lambda key=None: out.get(key) if key else out
+ return cpn, captured, out
+
+
+def _respond_with(monkeypatch, results):
+ monkeypatch.setattr(_keenable_mod, "_request", lambda *a, **k: {"results": results})
+
+
+def test_content_comes_from_snippet(monkeypatch):
+ # A realistic result: `description` empty, `snippet` carrying the page text.
+ _respond_with(
+ monkeypatch,
+ [{"title": "One", "url": "https://example.com/one", "description": "", "snippet": "First page text"}],
+ )
+ cpn, captured, _ = _build_tool()
+
+ cpn._invoke(query="anything")
+
+ assert captured["rendered"] == [{"title": "One", "url": "https://example.com/one", "content": "First page text"}]
+
+
+def test_content_falls_back_to_description(monkeypatch):
+ _respond_with(monkeypatch, [{"title": "One", "url": "https://example.com/one", "description": "A description"}])
+ cpn, captured, _ = _build_tool()
+
+ cpn._invoke(query="anything")
+
+ assert captured["rendered"][0]["content"] == "A description"
+
+
+def test_content_is_kept_whole_with_newlines_normalized(monkeypatch):
+ # Snippets arrive as raw page text. These chunks feed retrieval, so the text
+ # must not be truncated — only the newlines it arrives with are normalized.
+ body = "line one\n\nline two" + " padding" * 500
+ _respond_with(
+ monkeypatch,
+ [{"title": "One", "url": "https://example.com/one", "description": "", "snippet": body}],
+ )
+ cpn, captured, _ = _build_tool()
+
+ cpn._invoke(query="anything")
+
+ content = captured["rendered"][0]["content"]
+ assert "\n" not in content
+ assert content.startswith("line one line two")
+ assert content == " ".join(body.split())
+
+
+def test_missing_text_yields_empty_content(monkeypatch):
+ _respond_with(monkeypatch, [{"title": "One", "url": "https://example.com/one"}])
+ cpn, captured, _ = _build_tool()
+
+ cpn._invoke(query="anything")
+
+ assert captured["rendered"][0]["content"] == ""
+
+
+def test_blank_query_short_circuits():
+ cpn, _, out = _build_tool()
+
+ assert cpn._invoke(query="") == ""
+ assert out["formalized_content"] == ""