vamplabAI · virrius · Mar 16, 2026 · Feb 17, 2026 · Feb 17, 2026 · Feb 17, 2026
diff --git a/config.yaml.example b/config.yaml.example
@@ -57,15 +57,16 @@ tools:
     base_class: path.to.my.tools.CustomTool
   my_other_tool:
     base_class: "name_of_tool_class_in_registry"
-  # Search tools: configure Tavily API key and search limits per tool
+  # Search tools: configure search provider and API keys per tool
   # (can be overridden per-agent in tools list)
   web_search_tool:
-    tavily_api_key: "your-tavily-api-key-here"  # Tavily API key (get at tavily.com)
-    tavily_api_base_url: "https://api.tavily.com"  # Tavily API URL
+    engine: "tavily"  # Search engine: "tavily" (default), "brave", or "perplexity"
+    api_key: "your-search-api-key-here"  # API key for the selected engine
+    # api_base_url: "https://custom-url"  # Optional, uses engine default
     max_results: 12
     max_searches: 6
   extract_page_content_tool:
-    tavily_api_key: "your-tavily-api-key-here"  # Same Tavily API key
+    tavily_api_key: "your-tavily-api-key-here"  # Tavily API key (Tavily-only feature)
     tavily_api_base_url: "https://api.tavily.com"
     content_limit: 2000
 

diff --git a/examples/sgr_deep_research/config.yaml.example b/examples/sgr_deep_research/config.yaml.example
@@ -30,13 +30,12 @@ tools:
   # Core tools (base_class defaults to sgr_agent_core.tools.*)
   # Search tools: configure Tavily API key and search limits per tool
   web_search_tool:
-    tavily_api_key: "your-tavily-api-key-here"  # Tavily API key (get at tavily.com)
-    tavily_api_base_url: "https://api.tavily.com"  # Tavily API URL
+    engine: "tavily"  # Search engine: "tavily" (default), "brave", or "perplexity"
+    api_key: "your-tavily-api-key-here"  # API key for the selected engine
     max_searches: 4  # Max search operations
     max_results: 10  # Max results in search query
   extract_page_content_tool:
-    tavily_api_key: "your-tavily-api-key-here"  # Same Tavily API key
-    tavily_api_base_url: "https://api.tavily.com"
+    tavily_api_key: "your-tavily-api-key-here"  # Tavily API key (Tavily extract only)
     content_limit: 1500  # Content char limit per source
   create_report_tool:
     # base_class defaults to sgr_agent_core.tools.CreateReportTool

diff --git a/examples/sgr_deep_research_without_reporting/config.yaml.example b/examples/sgr_deep_research_without_reporting/config.yaml.example
@@ -29,13 +29,12 @@ tools:
   # Core tools (base_class defaults to sgr_agent_core.tools.*)
   # Search tools: configure Tavily API key and search limits per tool
   web_search_tool:
-    tavily_api_key: "your-tavily-api-key-here"  # Tavily API key (get at tavily.com)
-    tavily_api_base_url: "https://api.tavily.com"  # Tavily API URL
+    engine: "tavily"  # Search engine: "tavily" (default), "brave", or "perplexity"
+    api_key: "your-tavily-api-key-here"  # API key for the selected engine
     max_searches: 4  # Max search operations
     max_results: 10  # Max results in search query
   extract_page_content_tool:
-    tavily_api_key: "your-tavily-api-key-here"  # Same Tavily API key
-    tavily_api_base_url: "https://api.tavily.com"
+    tavily_api_key: "your-tavily-api-key-here"  # Tavily API key (Tavily extract only)
     content_limit: 1500  # Content char limit per source
   final_answer_tool:
     # base_class defaults to sgr_agent_core.tools.FinalAnswerTool

diff --git a/sgr_agent_core/__init__.py b/sgr_agent_core/__init__.py
@@ -30,7 +30,12 @@
     SourceData,
 )
 from sgr_agent_core.next_step_tool import NextStepToolsBuilder, NextStepToolStub
-from sgr_agent_core.services import AgentRegistry, MCP2ToolConverter, PromptLoader, ToolRegistry
+from sgr_agent_core.services import (
+    AgentRegistry,
+    MCP2ToolConverter,
+    PromptLoader,
+    ToolRegistry,
+)
 from sgr_agent_core.tools import *  # noqa: F403
 
 __all__ = [
@@ -50,9 +55,9 @@
     "SourceData",
     # Services
     "AgentRegistry",
-    "ToolRegistry",
-    "PromptLoader",
     "MCP2ToolConverter",
+    "PromptLoader",
+    "ToolRegistry",
     # Configuration
     "AgentConfig",
     "AgentDefinition",
@@ -64,11 +69,6 @@
     # Next step tools
     "NextStepToolStub",
     "NextStepToolsBuilder",
-    # Models
-    "AgentStatesEnum",
-    "AgentContext",
-    "SearchResult",
-    "SourceData",
     # Factory
     "AgentFactory",
 ]
diff --git a/sgr_agent_core/agent_definition.py b/sgr_agent_core/agent_definition.py
@@ -63,12 +63,11 @@ def to_openai_client_kwargs(self) -> dict[str, Any]:
 
 
 class SearchConfig(BaseModel, extra="allow"):
-    tavily_api_key: str | None = Field(default=None, description="Tavily API key")
-    tavily_api_base_url: str = Field(default="https://api.tavily.com", description="Tavily API base URL")
-
     max_searches: int = Field(default=4, ge=0, description="Maximum number of searches")
     max_results: int = Field(default=10, ge=1, description="Maximum number of search results")
     content_limit: int = Field(default=3500, gt=0, description="Content character limit per source")
+    tavily_api_key: str | None = Field(default=None, description="Tavily API key")
+    tavily_api_base_url: str = Field(default="https://api.tavily.com", description="Tavily API base URL")
 
 
 class PromptsConfig(BaseModel, extra="allow"):

diff --git a/sgr_agent_core/services/__init__.py b/sgr_agent_core/services/__init__.py
@@ -2,12 +2,14 @@
 
 from sgr_agent_core.services.mcp_service import MCP2ToolConverter
 from sgr_agent_core.services.prompt_loader import PromptLoader
-from sgr_agent_core.services.registry import AgentRegistry, StreamingGeneratorRegistry, ToolRegistry
-from sgr_agent_core.services.tavily_search import TavilySearchService
+from sgr_agent_core.services.registry import (
+    AgentRegistry,
+    StreamingGeneratorRegistry,
+    ToolRegistry,
+)
 from sgr_agent_core.services.tool_instantiator import ToolInstantiator
 
 __all__ = [
-    "TavilySearchService",
     "MCP2ToolConverter",
     "ToolRegistry",
     "StreamingGeneratorRegistry",

diff --git a/sgr_agent_core/services/tavily_search.py b/sgr_agent_core/services/tavily_search.py
diff --git a/sgr_agent_core/tools/__init__.py b/sgr_agent_core/tools/__init__.py
@@ -12,7 +12,7 @@
 from sgr_agent_core.tools.final_answer_tool import FinalAnswerTool
 from sgr_agent_core.tools.generate_plan_tool import GeneratePlanTool
 from sgr_agent_core.tools.reasoning_tool import ReasoningTool
-from sgr_agent_core.tools.web_search_tool import WebSearchTool
+from sgr_agent_core.tools.web_search_tool import WebSearchConfig, WebSearchTool
 
 __all__ = [
     # Base classes
@@ -24,16 +24,14 @@
     "ToolNameSelectorStub",
     "NextStepToolsBuilder",
     # Individual tools
-    "ClarificationTool",
-    "GeneratePlanTool",
-    "WebSearchTool",
-    "ExtractPageContentTool",
     "AdaptPlanTool",
-    "CreateReportTool",
     "AnswerTool",
+    "ClarificationTool",
+    "CreateReportTool",
+    "ExtractPageContentTool",
     "FinalAnswerTool",
+    "GeneratePlanTool",
     "ReasoningTool",
-    # Tool lists
-    "NextStepToolStub",
-    "NextStepToolsBuilder",
+    "WebSearchConfig",
+    "WebSearchTool",
 ]
diff --git a/sgr_agent_core/tools/extract_page_content_tool.py b/sgr_agent_core/tools/extract_page_content_tool.py
@@ -4,10 +4,11 @@
 from typing import TYPE_CHECKING, Any
 
 from pydantic import Field
+from tavily import AsyncTavilyClient
 
 from sgr_agent_core.agent_definition import SearchConfig
 from sgr_agent_core.base_tool import BaseTool
-from sgr_agent_core.services import TavilySearchService
+from sgr_agent_core.models import SourceData
 
 if TYPE_CHECKING:
     from sgr_agent_core.agent_definition import AgentConfig
@@ -19,11 +20,12 @@
 
 class ExtractPageContentTool(BaseTool):
     """Extract full detailed content from specific web pages.
-     Use for: Getting complete page content from URLs found in web search Returns:
-     Full page content in readable format (via Tavily Extract API)
-     Best for: Deep analysis of specific pages, extracting structured data
 
-    Usage: Call after WebSearchTool to get detailed information from promising URLs
+    Use for: Getting complete page content from URLs found in web search.
+    Returns: Full page content in readable format (via Tavily Extract API).
+    Best for: Deep analysis of specific pages, extracting structured data.
+
+    Usage: Call after WebSearchTool to get detailed information from promising URLs.
 
     CRITICAL WARNINGS:
         - Extracted pages may show data from DIFFERENT years/time periods than asked
@@ -38,13 +40,46 @@ class ExtractPageContentTool(BaseTool):
     reasoning: str = Field(description="Why extract these specific pages")
     urls: list[str] = Field(description="List of URLs to extract full content from", min_length=1, max_length=5)
 
+    @staticmethod
+    async def _extract(config: SearchConfig, urls: list[str]) -> list[SourceData]:
+        """Extract full content from URLs via Tavily Extract API."""
+        logger.info(f"Tavily extract: {len(urls)} URLs")
+
+        client = AsyncTavilyClient(api_key=config.tavily_api_key, api_base_url=config.tavily_api_base_url)
+        response = await client.extract(urls=urls)
+
+        sources = []
+        for i, result in enumerate(response.get("results", [])):
+            if not result.get("url"):
+                continue
+
+            source = SourceData(
+                number=i,
+                title=result.get("url", "").split("/")[-1] or "Extracted Content",
+                url=result.get("url", ""),
+                snippet="",
+                full_content=result.get("raw_content", ""),
+                char_count=len(result.get("raw_content", "")),
+            )
+            sources.append(source)
+
+        failed_urls = response.get("failed_results", [])
+        if failed_urls:
+            logger.warning(f"Failed to extract {len(failed_urls)} URLs: {failed_urls}")
+
+        return sources
+
     async def __call__(self, context: AgentContext, config: AgentConfig, **kwargs: Any) -> str:
         """Extract full content from specified URLs."""
         search_config = SearchConfig(**kwargs)
-        logger.info(f"📄 Extracting content from {len(self.urls)} URLs")
-
-        self._search_service = TavilySearchService(search_config)
-        sources = await self._search_service.extract(urls=self.urls)
+        if not search_config.tavily_api_key:
+            return (
+                "Error: tavily_api_key is required for ExtractPageContentTool."
+                " Tavily is the only provider that supports content extraction."
+            )
+        logger.info(f"Extracting content from {len(self.urls)} URLs")
+
+        sources = await self._extract(search_config, urls=self.urls)
 
         # Update existing sources instead of overwriting
         for source in sources: