diff --git a/python/configs/providers/openrouter.yaml b/python/configs/providers/openrouter.yaml
index 192b42251..b0660fc58 100644
--- a/python/configs/providers/openrouter.yaml
+++ b/python/configs/providers/openrouter.yaml
@@ -1,50 +1,62 @@
-  # ============================================
-  # OpenRouter Provider Configuration
-  # ============================================
-  name: "OpenRouter"
-  provider_type: "openrouter"
-
-  enabled: true # Default is true if not specified
-
-  # Connection Configuration
-  connection:
-    base_url: "https://openrouter.ai/api/v1"
-    api_key_env: "OPENROUTER_API_KEY"
-
-  # Default model if none specified
-  default_model: "anthropic/claude-haiku-4.5"
-
-  # Model Parameters Defaults
+# ============================================
+# OpenRouter Provider Configuration
+# ============================================
+name: "OpenRouter"
+provider_type: "openrouter"
+
+enabled: true # Default is true if not specified
+
+# Connection Configuration
+connection:
+  base_url: "https://openrouter.ai/api/v1"
+  api_key_env: "OPENROUTER_API_KEY"
+
+# Default model if none specified
+default_model: "anthropic/claude-haiku-4.5"
+
+# Model Parameters Defaults
+defaults:
+  temperature: 0.5
+
+# Extra headers for OpenRouter API
+extra_headers:
+  HTTP-Referer: "https://valuecell.ai"
+  X-Title: "ValueCell"
+
+# Available Models (commonly used)
+models:
+  - id: "anthropic/claude-haiku-4.5"
+    name: "Claude Haiku 4.5"
+  - id: "x-ai/grok-4"
+    name: "Grok 4"
+  - id: "qwen/qwen-max"
+    name: "Qwen Max"
+  - id: "openai/gpt-5"
+    name: "GPT-5"
+  - id: "google/gemini-2.5-flash"
+    name: "Gemini 2.5 Flash"
+  - id: "google/gemini-2.5-pro"
+    name: "Gemini 2.5 Pro"
+
+# ============================================
+# Embedding Models Configuration
+# ============================================
+# OpenRouter provides embedding models
+embedding:
+  # Default embedding model
+  default_model: "qwen/qwen3-embedding-4b"
+
+  # Default parameters
   defaults:
-    temperature: 0.5
-
-  # Extra headers for OpenRouter API
-  extra_headers:
-    HTTP-Referer: "https://valuecell.ai"
-    X-Title: "ValueCell"
+    dimensions: 2560
+    encoding_format: "float"
 
-  # Available Models (commonly used)
+  # Available embedding models
   models:
-
-    - id: "anthropic/claude-haiku-4.5"
-      name: "Claude Haiku 4.5"
-
-    - id: "x-ai/grok-4"
-      name: "Grok 4"
-
-    - id: "qwen/qwen-max"
-      name: "Qwen Max"
-
-    - id: "openai/gpt-5"
-      name: "GPT-5"
-
-    - id: "google/gemini-2.5-flash"
-      name: "Gemini 2.5 Flash"
-
-    - id: "google/gemini-2.5-pro"
-      name: "Gemini 2.5 Pro"
-
-  # Note: OpenRouter does not support embedding models
-  # For embedding, use other providers like OpenAI or SiliconFlow etc
+  - id: "qwen/qwen3-embedding-4b"
+    name: "Qwen3 Embedding 4B"
+    dimensions: 2560
+    max_input: 32768
+    description: "Qwen3 Embedding Model"
 
 
diff --git a/python/valuecell/agents/research_agent/core.py b/python/valuecell/agents/research_agent/core.py
index 1c83777a1..858f06bd9 100644
--- a/python/valuecell/agents/research_agent/core.py
+++ b/python/valuecell/agents/research_agent/core.py
@@ -7,7 +7,7 @@
 from loguru import logger
 
 import valuecell.utils.model as model_utils_mod
-from valuecell.agents.research_agent.knowledge import knowledge
+from valuecell.agents.research_agent.knowledge import get_knowledge
 from valuecell.agents.research_agent.prompts import (
     KNOWLEDGE_AGENT_EXPECTED_OUTPUT,
     KNOWLEDGE_AGENT_INSTRUCTION,
@@ -37,6 +37,8 @@ def __init__(self, **kwargs):
             # search_crypto_vcs,
             # search_crypto_people,
         ]
+        # Lazily obtain knowledge; disable search if unavailable
+        knowledge = get_knowledge()
         self.knowledge_research_agent = Agent(
             model=model_utils_mod.get_model_for_agent("research_agent"),
             instructions=[KNOWLEDGE_AGENT_INSTRUCTION],
@@ -45,7 +47,7 @@ def __init__(self, **kwargs):
             knowledge=knowledge,
             db=InMemoryDb(),
             # context
-            search_knowledge=True,
+            search_knowledge=knowledge is not None,
             add_datetime_to_context=True,
             add_history_to_context=True,
             num_history_runs=3,
@@ -54,7 +56,14 @@ def __init__(self, **kwargs):
             # configuration
             debug_mode=agent_debug_mode_enabled(),
         )
-        set_identity(os.getenv("SEC_EMAIL"))
+        # Configure EDGAR identity only when SEC_EMAIL is present
+        sec_email = os.getenv("SEC_EMAIL")
+        if sec_email:
+            set_identity(sec_email)
+        else:
+            logger.warning(
+                "SEC_EMAIL not set; EDGAR identity is not configured for ResearchAgent."
+            )
 
     async def stream(
         self,
diff --git a/python/valuecell/agents/research_agent/knowledge.py b/python/valuecell/agents/research_agent/knowledge.py
index 8be8447ff..82d6bd8bc 100644
--- a/python/valuecell/agents/research_agent/knowledge.py
+++ b/python/valuecell/agents/research_agent/knowledge.py
@@ -5,13 +5,44 @@
 from agno.knowledge.knowledge import Knowledge
 from agno.knowledge.reader.markdown_reader import MarkdownReader
 from agno.knowledge.reader.pdf_reader import PDFReader
+from loguru import logger
+
+from .vdb import get_vector_db
+
+_knowledge_cache: Optional[Knowledge] = None
+
+
+def get_knowledge() -> Optional[Knowledge]:
+    """Lazily create and cache the Knowledge instance.
+
+    Returns None when embeddings/vector DB are unavailable, enabling a
+    tools-only mode without knowledge search.
+    """
+    global _knowledge_cache
+    if _knowledge_cache is not None:
+        return _knowledge_cache
+
+    vdb = get_vector_db()
+    if vdb is None:
+        logger.warning(
+            "ResearchAgent Knowledge disabled: vector DB unavailable (no embeddings)."
+        )
+        return None
+
+    try:
+        _knowledge_cache = Knowledge(
+            vector_db=vdb,
+            max_results=10,
+        )
+        return _knowledge_cache
+    except Exception as e:
+        logger.warning(
+            "Failed to create Knowledge for ResearchAgent; disabling. Error: {}",
+            e,
+        )
+        return None
 
-from .vdb import vector_db
 
-knowledge = Knowledge(
-    vector_db=vector_db,
-    max_results=10,
-)
 md_reader = MarkdownReader(chunking_strategy=MarkdownChunking())
 pdf_reader = PDFReader(chunking_strategy=MarkdownChunking())
 
@@ -19,6 +50,12 @@
 async def insert_md_file_to_knowledge(
     name: str, path: Path, metadata: Optional[dict] = None
 ):
+    knowledge = get_knowledge()
+    if knowledge is None:
+        logger.warning(
+            "Skipping markdown insertion: Knowledge disabled (no embeddings configured)."
+        )
+        return
     await knowledge.add_content_async(
         name=name,
         path=path,
@@ -28,6 +65,12 @@ async def insert_md_file_to_knowledge(
 
 
 async def insert_pdf_file_to_knowledge(url: str, metadata: Optional[dict] = None):
+    knowledge = get_knowledge()
+    if knowledge is None:
+        logger.warning(
+            "Skipping PDF insertion: Knowledge disabled (no embeddings configured)."
+        )
+        return
     await knowledge.add_content_async(
         url=url,
         metadata=metadata,
diff --git a/python/valuecell/agents/research_agent/tests/test_fault_tolerance.py b/python/valuecell/agents/research_agent/tests/test_fault_tolerance.py
new file mode 100644
index 000000000..9dd5e2809
--- /dev/null
+++ b/python/valuecell/agents/research_agent/tests/test_fault_tolerance.py
@@ -0,0 +1,230 @@
+import pytest
+
+
+def _raise_unavailable(*args, **kwargs):
+    raise ValueError("No embedding provider configured")
+
+
+def test_get_vector_db_returns_none_when_embedder_unavailable(monkeypatch):
+    # Cause embedder creation to fail
+    import valuecell.agents.research_agent.vdb as vdb
+
+    monkeypatch.setattr(
+        vdb.model_utils_mod, "get_embedder_for_agent", _raise_unavailable
+    )
+
+    assert vdb.get_vector_db() is None
+
+
+def test_get_knowledge_returns_none_without_embeddings(monkeypatch):
+    # Cause embedder creation to fail
+    import valuecell.agents.research_agent.vdb as vdb
+
+    monkeypatch.setattr(
+        vdb.model_utils_mod, "get_embedder_for_agent", _raise_unavailable
+    )
+
+    from valuecell.agents.research_agent.knowledge import get_knowledge
+
+    assert get_knowledge() is None
+
+
+def test_research_agent_initializes_without_knowledge_when_embeddings_missing(
+    monkeypatch,
+):
+    # Cause embedder creation to fail
+    import valuecell.agents.research_agent.vdb as vdb
+
+    monkeypatch.setattr(
+        vdb.model_utils_mod, "get_embedder_for_agent", _raise_unavailable
+    )
+
+    # Stub model creation to avoid provider requirements
+    import valuecell.utils.model as model_utils_mod
+
+    monkeypatch.setattr(model_utils_mod, "get_model_for_agent", lambda name: object())
+
+    # Replace Agent with a dummy capturing params
+    import valuecell.agents.research_agent.core as core_mod
+
+    class DummyAgent:
+        def __init__(
+            self,
+            *,
+            model,
+            instructions,
+            expected_output,
+            tools,
+            knowledge,
+            db,
+            search_knowledge,
+            **kwargs,
+        ):
+            self.model = model
+            self.instructions = instructions
+            self.expected_output = expected_output
+            self.tools = tools
+            self.knowledge = knowledge
+            self.db = db
+            self.search_knowledge = search_knowledge
+
+    monkeypatch.setattr(core_mod, "Agent", DummyAgent)
+
+    from valuecell.agents.research_agent.core import ResearchAgent
+
+    ra = ResearchAgent()
+    assert ra.knowledge_research_agent.knowledge is None
+    assert ra.knowledge_research_agent.search_knowledge is False
+
+
+def test_get_vector_db_success_path(monkeypatch):
+    import valuecell.agents.research_agent.vdb as vdb
+
+    # Provide a fake embedder and LanceDb to exercise success path
+    monkeypatch.setattr(
+        vdb.model_utils_mod, "get_embedder_for_agent", lambda name: object()
+    )
+
+    class DummyLanceDb:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+
+    monkeypatch.setattr(vdb, "LanceDb", DummyLanceDb)
+
+    db = vdb.get_vector_db()
+    assert isinstance(db, DummyLanceDb)
+
+    # Exercise failure path in LanceDb constructor to cover exception branch
+    class RaisingLanceDb:
+        def __init__(self, **kwargs):
+            raise RuntimeError("boom")
+
+    monkeypatch.setattr(vdb, "LanceDb", RaisingLanceDb)
+    assert vdb.get_vector_db() is None
+
+
+@pytest.mark.asyncio
+async def test_insert_functions_noop_when_disabled(monkeypatch, tmp_path):
+    # Cause embedder creation to fail
+    import valuecell.agents.research_agent.vdb as vdb
+
+    monkeypatch.setattr(
+        vdb.model_utils_mod, "get_embedder_for_agent", _raise_unavailable
+    )
+
+    from valuecell.agents.research_agent.knowledge import (
+        insert_md_file_to_knowledge,
+        insert_pdf_file_to_knowledge,
+    )
+
+    md_file = tmp_path / "doc.md"
+    md_file.write_text("# Title\nBody")
+
+    # Should not raise even though knowledge is disabled
+    await insert_md_file_to_knowledge("doc", md_file)
+    await insert_pdf_file_to_knowledge("https://example.com/doc.pdf")
+
+
+@pytest.mark.asyncio
+async def test_insert_functions_invoke_add_content_when_enabled(monkeypatch, tmp_path):
+    # Patch get_knowledge to return a dummy knowledge object
+    from valuecell.agents.research_agent import knowledge as knowledge_mod
+
+    class DummyKnowledge:
+        def __init__(self):
+            self.calls = []
+
+        async def add_content_async(self, **kwargs):
+            self.calls.append(kwargs)
+
+    dummy = DummyKnowledge()
+    monkeypatch.setattr(knowledge_mod, "_knowledge_cache", None)
+    monkeypatch.setattr(knowledge_mod, "get_knowledge", lambda: dummy)
+
+    from valuecell.agents.research_agent.knowledge import (
+        insert_md_file_to_knowledge,
+        insert_pdf_file_to_knowledge,
+    )
+
+    md_file = tmp_path / "doc2.md"
+    md_file.write_text("# Title\nBody")
+
+    await insert_md_file_to_knowledge("doc2", md_file, metadata={"k": "v"})
+    await insert_pdf_file_to_knowledge("https://example.com/doc2.pdf")
+
+    # Verify that add_content_async was called for both inserts
+    assert len(dummy.calls) == 2
+    assert any("path" in c for c in dummy.calls)
+    assert any("url" in c for c in dummy.calls)
+
+
+def test_get_knowledge_success_path_creates_and_caches(monkeypatch):
+    from valuecell.agents.research_agent import knowledge as knowledge_mod
+
+    # Use a dummy Knowledge to avoid external dependency behavior
+    class DummyKnowledge:
+        def __init__(self, *, vector_db, max_results):
+            self.vector_db = vector_db
+            self.max_results = max_results
+
+    monkeypatch.setattr(knowledge_mod, "Knowledge", DummyKnowledge)
+    monkeypatch.setattr(knowledge_mod, "_knowledge_cache", None)
+    monkeypatch.setattr(knowledge_mod, "get_vector_db", lambda: object())
+
+    from valuecell.agents.research_agent.knowledge import get_knowledge
+
+    k1 = get_knowledge()
+    k2 = get_knowledge()
+    assert isinstance(k1, DummyKnowledge)
+    assert k1 is k2  # cached
+
+
+@pytest.mark.asyncio
+async def test_stream_yields_events_without_knowledge(monkeypatch):
+    # Cause embedder creation to fail and stub model creation
+    import valuecell.agents.research_agent.vdb as vdb
+
+    monkeypatch.setattr(
+        vdb.model_utils_mod, "get_embedder_for_agent", _raise_unavailable
+    )
+
+    import valuecell.utils.model as model_utils_mod
+
+    monkeypatch.setattr(model_utils_mod, "get_model_for_agent", lambda name: object())
+
+    import types as _types
+    import valuecell.agents.research_agent.core as core_mod
+
+    class DummyAgent:
+        def __init__(self, **kwargs):
+            pass
+
+        async def arun(self, *args, **kwargs):
+            # Yield three events to exercise stream handling
+            yield _types.SimpleNamespace(event="RunContent", content="hello")
+            yield _types.SimpleNamespace(
+                event="ToolCallStarted",
+                tool=_types.SimpleNamespace(tool_call_id="id1", tool_name="foo"),
+            )
+            yield _types.SimpleNamespace(
+                event="ToolCallCompleted",
+                tool=_types.SimpleNamespace(
+                    result="ok", tool_call_id="id1", tool_name="foo"
+                ),
+            )
+
+    monkeypatch.setattr(core_mod, "Agent", DummyAgent)
+    from valuecell.agents.research_agent.core import ResearchAgent
+
+    ra = ResearchAgent()
+    # Iterate the stream to ensure branches are executed
+    events = []
+    async for ev in ra.stream(
+        query="q",
+        conversation_id="c",
+        task_id="t",
+        dependencies={"a": 1},
+    ):
+        events.append(ev)
+    # Should have yielded message chunk, tool start, tool complete, and done
+    assert len(events) == 4
diff --git a/python/valuecell/agents/research_agent/vdb.py b/python/valuecell/agents/research_agent/vdb.py
index 4aa04f4ac..f60b9b480 100644
--- a/python/valuecell/agents/research_agent/vdb.py
+++ b/python/valuecell/agents/research_agent/vdb.py
@@ -1,44 +1,53 @@
 """
 Vector database configuration for Research Agent.
 
-This module uses the centralized configuration system to create an embedder
-and vector database. It automatically:
-1. Selects an available provider with embedding support (OpenAI, SiliconFlow, etc.)
-2. Uses the provider's API key from .env
-3. Falls back to other providers if the primary fails
-4. Respects environment variable overrides (EMBEDDER_MODEL_ID, EMBEDDER_DIMENSION)
-
-Configuration Priority (highest to lowest):
-1. Environment Variables (EMBEDDER_MODEL_ID, EMBEDDER_DIMENSION, etc.)
-2. .env file (OPENROUTER_API_KEY, SILICONFLOW_API_KEY, etc.)
-3. YAML files (configs/agents/research_agent.yaml, configs/providers/*.yaml)
+Fault-tolerant, lazy initialization:
+- Attempts to create an embedder and LanceDb only when requested.
+- If no embedding provider/API key is available, returns None instead of raising.
+- Respects environment variable overrides (e.g., EMBEDDER_MODEL_ID).
+
+This prevents import-time failures and allows the ResearchAgent to run in a
+"tools-only" mode without knowledge search when embeddings are not configured.
 """
 
+from typing import Optional
+
 from agno.vectordb.lancedb import LanceDb
 from agno.vectordb.search import SearchType
+from loguru import logger
 
 import valuecell.utils.model as model_utils_mod
 from valuecell.utils.db import resolve_lancedb_uri
 
-# Create embedder using the configuration system
-# This will:
-# - Check EMBEDDER_MODEL_ID env var first
-# - Auto-select provider with embedding support (e.g., SiliconFlow if SILICONFLOW_API_KEY is set)
-# - Use provider's default embedding model if not specified
-# - Fall back to other providers if primary fails
-embedder = model_utils_mod.get_embedder_for_agent("research_agent")
-
-# Alternative usage examples:
-# embedder = get_embedder()  # Use default env key
-# embedder = get_embedder("EMBEDDER_MODEL_ID", dimensions=3072)  # Override dimensions
-# embedder = get_embedder_for_agent("research_agent")  # Use agent-specific config
-
-# Create vector database with the configured embedder
-vector_db = LanceDb(
-    table_name="research_agent_knowledge_base",
-    uri=resolve_lancedb_uri(),
-    embedder=embedder,
-    # reranker=reranker,  # Optional: can be configured later, reranker config in modelprovider yaml file if needed
-    search_type=SearchType.hybrid,
-    use_tantivy=False,
-)
+
+def get_vector_db() -> Optional[LanceDb]:
+    """Create and return the LanceDb instance, or None if embeddings are unavailable.
+
+    This function is safe to call at runtime; it will not raise during normal
+    missing-configuration scenarios. Unexpected errors are logged and result in
+    a None return to enable graceful degradation.
+    """
+    try:
+        embedder = model_utils_mod.get_embedder_for_agent("research_agent")
+    except Exception as e:
+        logger.warning(
+            "ResearchAgent embeddings unavailable; disabling knowledge search. Error: {}",
+            e,
+        )
+        return None
+
+    try:
+        return LanceDb(
+            table_name="research_agent_knowledge_base",
+            uri=resolve_lancedb_uri(),
+            embedder=embedder,
+            # reranker=reranker,  # Optional: can be configured later
+            search_type=SearchType.hybrid,
+            use_tantivy=False,
+        )
+    except Exception as e:
+        logger.warning(
+            "Failed to initialize LanceDb for ResearchAgent; disabling knowledge. Error: {}",
+            e,
+        )
+        return None