fix(research): support timeout defaults in direct tests (#2624)

fix(research): honor planning query timeouts

src/deep_research.py

@@ -196,6 +196,8 @@ class DeepResearcher:
         max_content_chars: int = 15000,
         max_report_tokens: int = 8192,
         extraction_timeout: int = 90,
+        planning_timeout: int = 90,
+        query_timeout: int = 120,
         extraction_concurrency: int = 3,
         min_rounds: int = 2,
         max_empty_rounds: int = 2,
@@ -215,6 +217,8 @@ class DeepResearcher:
         self.max_content_chars = max_content_chars
         self.max_report_tokens = max_report_tokens
         self.extraction_timeout = min(3600, max(15, int(extraction_timeout or 90)))
+        self.planning_timeout = min(3600, max(15, int(planning_timeout or 90)))
+        self.query_timeout = min(3600, max(15, int(query_timeout or 120)))
         self.extraction_concurrency = min(12, max(1, int(extraction_concurrency or 3)))
         self.min_rounds = min_rounds
         self.max_empty_rounds = max_empty_rounds
@@ -395,7 +399,7 @@ class DeepResearcher:
                 [{"role": "user", "content": prompt}],
                 temperature=0.3,
                 max_tokens=1024,
-                timeout=30,
+                timeout=getattr(self, "planning_timeout", 90),
             )
             # Try to parse as JSON for structured plan
             parsed = self._parse_json_object(response)
@@ -478,6 +482,7 @@ class DeepResearcher:
                 [{"role": "user", "content": prompt}],
                 temperature=0.5,
                 max_tokens=4096,
+                timeout=getattr(self, "query_timeout", 120),
             )
             queries = self._parse_json_array(response)
             # Deduplicate

src/research_handler.py

@@ -722,6 +722,18 @@ class ResearchHandler:
                 minimum=1,
                 maximum=12,
             )
+            _planning_timeout = _bounded_int(
+                get_setting("research_planning_timeout_seconds", _extraction_timeout),
+                default=_extraction_timeout,
+                minimum=15,
+                maximum=3600,
+            )
+            _query_timeout = _bounded_int(
+                get_setting("research_query_timeout_seconds", _extraction_timeout),
+                default=_extraction_timeout,
+                minimum=15,
+                maximum=3600,
+            )
 
             researcher = DeepResearcher(
                 llm_endpoint=llm_endpoint,
@@ -732,6 +744,8 @@ class ResearchHandler:
                 max_time=max_time,
                 max_report_tokens=_max_report_tokens,
                 extraction_timeout=_extraction_timeout,
+                planning_timeout=_planning_timeout,
+                query_timeout=_query_timeout,
                 extraction_concurrency=_extraction_concurrency,
                 progress_callback=progress_callback,
                 search_provider=search_provider,

src/settings.py

@@ -85,6 +85,11 @@ DEFAULT_SETTINGS = {
     "research_search_provider": "",
     "research_max_tokens": 16384,
     "research_extraction_timeout_seconds": 90,
+    # Lightweight planning/query LLM calls happen before any search starts.
+    # Keep them separately tunable so slow local backends are not capped by
+    # the old 30s/60s per-call defaults.
+    "research_planning_timeout_seconds": 90,
+    "research_query_timeout_seconds": 90,
     "research_extraction_concurrency": 3,
     # Hard wall-clock cap on a single deep-research run. The previous 600s
     # (10 min) default cut off slow local / edge LLMs mid-synthesis; 1800s

tests/test_deep_research_extraction_controls.py

@@ -96,3 +96,33 @@ def test_extraction_timeout_allows_long_local_model_runs():
     )
 
     assert researcher.extraction_timeout == 1800
+
+
+@pytest.mark.asyncio
+async def test_planning_and_query_generation_use_configured_timeouts():
+    researcher = DeepResearcher(
+        llm_endpoint="http://local.test/v1/chat/completions",
+        llm_model="local-model",
+        planning_timeout=234,
+        query_timeout=345,
+    )
+    captured = []
+
+    async def fake_llm(messages, temperature=0.3, max_tokens=4096, timeout=60):
+        captured.append(timeout)
+        if max_tokens == 1024:
+            return json.dumps({
+                "sub_questions": ["one"],
+                "key_topics": ["topic"],
+                "success_criteria": "complete",
+            })
+        return json.dumps(["query one", "query two"])
+
+    researcher._llm = fake_llm
+
+    plan = await researcher._create_plan("question")
+    queries = await researcher._generate_queries("question", "", 1)
+
+    assert "Sub-questions: one" in plan
+    assert queries == ["query one", "query two"]
+    assert captured == [234, 345]