Multi-provider search (Brave, Perplexity) (#177)

* feat(search): add multi-provider search services with BaseSearchService, Brave and Perplexity

Add BaseSearchService base class with factory pattern routing by engine config.
Add BraveSearchService (httpx, native offset) and PerplexitySearchService (httpx, over-fetch+slice).
Extend SearchConfig with engine, brave_*, perplexity_* fields. Refactor TavilySearchService to inherit BaseSearchService.

* feat(search): add standalone search tools (Tavily, Brave, Perplexity) with shared BaseSearchTool base

Add _BaseSearchTool base class with shared fields and __call__ logic
Refactor WebSearchTool to inherit from _BaseSearchTool
Add TavilySearchTool, BraveSearchTool, PerplexitySearchTool standalone tools
Fix description ClassVar inheritance in __init_subclass__
Add Tavily key guard to ExtractPageContentTool
Update exports and config examples with multi-provider settings

* test(search): add search services tests and update tool tests for multi-provider support

Add test_search_services.py with factory, conversion and missing API key tests
Update test_tools.py: patch BaseSearchService instead of TavilySearchService, add init/execution tests for BraveSearchTool, PerplexitySearchTool, TavilySearchTool

* refactor(exports): reorder and update exports to include TavilySearchService and adjust tool exports for consistency

* docs(tavily_search): add class docstring explaining usage, auth, and offset handling

* refactor(search): remove deprecated BaseSearchService and inline provider logic into tools with unified registry and tests

* refactor(search): replace _search_registry with SearchProviderRegistry for search tools to centralize and standardize search provider registration and lookup

* refactor(extract_page_content_tool): move Tavily extract logic from tavily_search_tool to improve separation of concerns and update tests accordingly

* refactor(agent_definition): reorganize SearchConfig fields by provider and add section comments for clarity

* docs(web_search_tool): expand class docstring to clarify engine selection and usage options

* refactor(search): consolidate search providers into unified web search tool

- consolidate search providers into a unified web search tool and config
- remove SearchProviderRegistry and legacy base_search_tool and provider modules
- update example configs to use generic api_key and optional api_base_url
- adjust exports and registries to reflect removed search-specific classes

* refactor(tools): add ExtractPageContentConfig and refactor extraction tool

- extract SearchConfig into ExtractPageContentConfig in tools
- update ExtractPageContentTool to use new config and validate tavily_api_key
- adjust package exports and imports to expose new config class
- update docs/comments and tests to reference ExtractPageContentConfig

Author: nikmd1306

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
 

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

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

sgr_agent_core/__init__.py

@@ -16,7 +16,6 @@
     ExecutionConfig,
     LLMConfig,
     PromptsConfig,
-    SearchConfig,
 )
 from sgr_agent_core.agent_factory import AgentFactory
 from sgr_agent_core.agents import *  # noqa: F403
@@ -30,7 +29,12 @@
     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__ = [
@@ -50,25 +54,19 @@
     "SourceData",
     # Services
     "AgentRegistry",
-    "ToolRegistry",
-    "PromptLoader",
     "MCP2ToolConverter",
+    "PromptLoader",
+    "ToolRegistry",
     # Configuration
     "AgentConfig",
     "AgentDefinition",
     "LLMConfig",
     "PromptsConfig",
-    "SearchConfig",
     "ExecutionConfig",
     "GlobalConfig",
     # Next step tools
     "NextStepToolStub",
     "NextStepToolsBuilder",
-    # Models
-    "AgentStatesEnum",
-    "AgentContext",
-    "SearchResult",
-    "SourceData",
     # Factory
     "AgentFactory",
 ]

sgr_agent_core/agent_definition.py

@@ -62,15 +62,6 @@ def to_openai_client_kwargs(self) -> dict[str, Any]:
         return self.model_dump(exclude={"api_key", "base_url", "proxy"})
 
 
-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")
-
-
 class PromptsConfig(BaseModel, extra="allow"):
     system_prompt_file: FilePath | None = Field(
         default=os.path.join(os.path.dirname(__file__), "prompts/system_prompt.txt"),

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",

sgr_agent_core/services/tavily_search.py

@@ -8,11 +8,11 @@
 from sgr_agent_core.tools.answer_tool import AnswerTool
 from sgr_agent_core.tools.clarification_tool import ClarificationTool
 from sgr_agent_core.tools.create_report_tool import CreateReportTool
-from sgr_agent_core.tools.extract_page_content_tool import ExtractPageContentTool
+from sgr_agent_core.tools.extract_page_content_tool import ExtractPageContentConfig, ExtractPageContentTool
 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 +24,15 @@
     "ToolNameSelectorStub",
     "NextStepToolsBuilder",
     # Individual tools
-    "ClarificationTool",
-    "GeneratePlanTool",
-    "WebSearchTool",
-    "ExtractPageContentTool",
     "AdaptPlanTool",
-    "CreateReportTool",
     "AnswerTool",
+    "ClarificationTool",
+    "CreateReportTool",
+    "ExtractPageContentConfig",
+    "ExtractPageContentTool",
     "FinalAnswerTool",
+    "GeneratePlanTool",
     "ReasoningTool",
-    # Tool lists
-    "NextStepToolStub",
-    "NextStepToolsBuilder",
+    "WebSearchConfig",
+    "WebSearchTool",
 ]