local-deep-research 0.2.2__py3-none-any.whl → 0.3.0__py3-none-any.whl

local_deep_research/__init__.py

@@ -2,7 +2,7 @@
 Local Deep Research - A tool for conducting deep research using AI.
 """
 
-__version__ = "0.1.0"
+__version__ = "0.2.0"
 __author__ = "Your Name"
 __description__ = "A tool for conducting deep research using AI"
 

local_deep_research/advanced_search_system/filters/cross_engine_filter.py

@@ -6,6 +6,7 @@ import json
 import logging
 from typing import Dict, List
 
+from ...utilities.db_utils import get_db_setting
 from ...utilities.search_utilities import remove_think_tags
 from .base_filter import BaseFilter
 
@@ -16,7 +17,7 @@ class CrossEngineFilter(BaseFilter):
     """Filter that ranks and filters results from multiple search engines."""
 
     def __init__(
-        self, model, max_results=20, default_reorder=True, default_reindex=True
+        self, model, max_results=None, default_reorder=True, default_reindex=True
     ):
         """
         Initialize the cross-engine filter.
@@ -28,6 +29,9 @@ class CrossEngineFilter(BaseFilter):
             default_reindex: Default setting for reindexing results after filtering
         """
         super().__init__(model)
+        # Get max_results from database settings if not provided
+        if max_results is None:
+            max_results = get_db_setting("search.cross_engine_max_results", 100)
         self.max_results = max_results
         self.default_reorder = default_reorder
         self.default_reindex = default_reindex

local_deep_research/advanced_search_system/strategies/base_strategy.py

@@ -13,11 +13,14 @@ logger = logging.getLogger(__name__)
 class BaseSearchStrategy(ABC):
     """Abstract base class for all search strategies."""
 
-    def __init__(self):
+    def __init__(self, all_links_of_system=None):
         """Initialize the base strategy with common attributes."""
         self.progress_callback = None
         self.questions_by_iteration = {}
-        self.all_links_of_system = []
+        # Create a new list if None is provided (avoiding mutable default argument)
+        self.all_links_of_system = (
+            all_links_of_system if all_links_of_system is not None else []
+        )
 
     def set_progress_callback(self, callback: Callable[[str, int, dict], None]) -> None:
         """Set a callback function to receive progress updates."""

local_deep_research/advanced_search_system/strategies/iterdrag_strategy.py

@@ -7,10 +7,7 @@ import logging
 from datetime import datetime
 from typing import Dict, List
 
-from langchain_core.language_models import BaseLLM
-
 from ...citation_handler import CitationHandler
-from ...config.config_files import settings
 from ...config.llm_config import get_llm
 from ...config.search_config import get_search
 from ...utilities.db_utils import get_db_setting
@@ -27,18 +24,34 @@ class IterDRAGStrategy(BaseSearchStrategy):
     """IterDRAG strategy that breaks queries into sub-queries."""
 
     def __init__(
-        self, model: BaseLLM | None = None, search=None, citation_handler=None
+        self,
+        search=None,
+        model=None,
+        max_iterations=3,
+        subqueries_per_iteration=2,
+        all_links_of_system=None,
     ):
-        """Initialize the strategy with optional dependency injection for testing."""
-        super().__init__()
-        self.model = model or get_llm()
+        """Initialize the IterDRAG strategy with search and LLM.
+
+        Args:
+            search: Search engine to use for web queries
+            model: LLM to use for text generation and reasoning
+            max_iterations: Maximum number of iterations to run
+            subqueries_per_iteration: Number of sub-queries to generate per iteration
+            all_links_of_system: Optional list of links to initialize with
+        """
+        super().__init__(all_links_of_system=all_links_of_system)
         self.search = search or get_search()
+        self.model = model or get_llm()
+        self.max_iterations = max_iterations
+        self.subqueries_per_iteration = subqueries_per_iteration
+
+        # Initialize progress callback
         self.progress_callback = None
-        self.all_links_of_system = list()
         self.questions_by_iteration = {}
 
         # Use provided citation_handler or create one
-        self.citation_handler = citation_handler or CitationHandler(self.model)
+        self.citation_handler = CitationHandler(self.model)
 
         # Initialize components
         self.question_generator = DecompositionQuestionGenerator(self.model)
@@ -396,13 +409,7 @@ Please try again with a different query or contact support.
                     """
 
         # Compress knowledge if needed
-        if (
-            get_db_setting(
-                "general.knowledge_accumulation",
-                settings.general.knowledge_accumulation,
-            )
-            == "ITERATION"
-        ):
+        if get_db_setting("general.knowledge_accumulation", "ITERATION") == "ITERATION":
             try:
                 self._update_progress(
                     "Compressing knowledge", 90, {"phase": "knowledge_compression"}

local_deep_research/advanced_search_system/strategies/parallel_search_strategy.py

@@ -34,7 +34,8 @@ class ParallelSearchStrategy(BaseSearchStrategy):
         use_cross_engine_filter: bool = True,
         filter_reorder: bool = True,
         filter_reindex: bool = True,
-        filter_max_results: int = 20,
+        cross_engine_max_results: int = None,
+        all_links_of_system=None,
     ):
         """Initialize with optional dependency injection for testing.
 
@@ -46,23 +47,29 @@ class ParallelSearchStrategy(BaseSearchStrategy):
             use_cross_engine_filter: If True, filter search results across engines
             filter_reorder: Whether to reorder results by relevance
             filter_reindex: Whether to update result indices after filtering
-            filter_max_results: Maximum number of results to keep after filtering
+            cross_engine_max_results: Maximum number of results to keep after cross-engine filtering
+            all_links_of_system: Optional list of links to initialize with
         """
-        super().__init__()
+        super().__init__(all_links_of_system=all_links_of_system)
         self.search = search or get_search()
         self.model = model or get_llm()
         self.progress_callback = None
-        self.all_links_of_system = list()
         self.questions_by_iteration = {}
         self.include_text_content = include_text_content
         self.use_cross_engine_filter = use_cross_engine_filter
         self.filter_reorder = filter_reorder
         self.filter_reindex = filter_reindex
 
+        # Get max_filtered_results from database if not provided
+        if cross_engine_max_results is None:
+            cross_engine_max_results = get_db_setting(
+                "search.cross_engine_max_results", 100
+            )
+
         # Initialize the cross-engine filter
         self.cross_engine_filter = CrossEngineFilter(
             model=self.model,
-            max_results=filter_max_results,
+            max_results=cross_engine_max_results,
             default_reorder=filter_reorder,
             default_reindex=filter_reindex,
         )
@@ -80,7 +87,7 @@ class ParallelSearchStrategy(BaseSearchStrategy):
 
     def analyze_topic(self, query: str) -> Dict:
         """
-        Parallel implementation that generates questions and searches all at once.
+        Analyze a topic using parallel search, supporting multiple iterations.
 
         Args:
             query: The research query to analyze
@@ -89,6 +96,11 @@ class ParallelSearchStrategy(BaseSearchStrategy):
 
         findings = []
         all_search_results = []
+        current_knowledge = ""
+
+        # Track all search results across iterations
+        self.all_links_of_system = list()
+        self.questions_by_iteration = {}
 
         self._update_progress(
             "Initializing parallel research",
@@ -105,184 +117,301 @@ class ParallelSearchStrategy(BaseSearchStrategy):
             return {
                 "findings": [],
                 "iterations": 0,
-                "questions": {},
+                "questions_by_iteration": {},
                 "formatted_findings": "Error: Unable to conduct research without a search engine.",
                 "current_knowledge": "",
                 "error": "No search engine available",
             }
 
+        # Determine number of iterations to run
+        iterations_to_run = get_db_setting("search.iterations")
+        logger.debug("Selected amount of iterations: " + str(iterations_to_run))
+        iterations_to_run = int(iterations_to_run)
         try:
-            # Step 1: Generate questions first
-            self._update_progress(
-                "Generating search questions", 10, {"phase": "question_generation"}
-            )
+            # Run each iteration
+            for iteration in range(1, iterations_to_run + 1):
+                iteration_progress_base = 5 + (iteration - 1) * (70 / iterations_to_run)
 
-            # Generate 3 additional questions (plus the main query = 4 total)
-            questions = self.question_generator.generate_questions(
-                current_knowledge="",  # No knowledge accumulation
-                query=query,
-                questions_per_iteration=int(
-                    get_db_setting("search.questions_per_iteration")
-                ),  # 3 additional questions
-                questions_by_iteration={},
-            )
+                self._update_progress(
+                    f"Starting iteration {iteration}/{iterations_to_run}",
+                    iteration_progress_base,
+                    {"phase": f"iteration_{iteration}", "iteration": iteration},
+                )
+
+                # Step 1: Generate questions
+                self._update_progress(
+                    f"Generating search questions for iteration {iteration}",
+                    iteration_progress_base + 5,
+                    {"phase": "question_generation", "iteration": iteration},
+                )
 
-            # Add the original query as the first question
-            all_questions = [query] + questions
+                # For first iteration, generate initial questions
+                # For subsequent iterations, generate follow-up questions
+                logger.info("Starting to generate questions")
+                if iteration == 1:
+                    # Generate additional questions (plus the main query)
+                    if iterations_to_run > 1:
+                        context = f"""Iteration: {1} of {iterations_to_run}"""
+                    else:
+                        context = ""
+                    questions = self.question_generator.generate_questions(
+                        current_knowledge=context,
+                        query=query,
+                        questions_per_iteration=int(
+                            get_db_setting("search.questions_per_iteration")
+                        ),
+                        questions_by_iteration=self.questions_by_iteration,
+                    )
 
-            # Store in questions_by_iteration
-            self.questions_by_iteration[0] = questions
-            logger.info(f"Generated questions: {questions}")
+                    # Add the original query as the first question
+                    all_questions = [query] + questions
 
-            # Step 2: Run all searches in parallel
-            self._update_progress(
-                "Running parallel searches for all questions",
-                20,
-                {"phase": "parallel_search"},
-            )
+                    # Store in questions_by_iteration
+                    self.questions_by_iteration[iteration] = questions
+                    logger.info(
+                        f"Generated questions for iteration {iteration}: {questions}"
+                    )
+                else:
+                    # Get past questions from all previous iterations
+                    past_questions = []
+                    for prev_iter in range(1, iteration):
+                        if prev_iter in self.questions_by_iteration:
+                            past_questions.extend(
+                                self.questions_by_iteration[prev_iter]
+                            )
+
+                    # Generate follow-up questions based on accumulated knowledge if iterations > 2
+                    use_knowledge = iterations_to_run > 2
+                    knowledge_for_questions = current_knowledge if use_knowledge else ""
+                    context = f"""Current Knowledge: {knowledge_for_questions}
+                    Iteration: {iteration} of {iterations_to_run}"""
+
+                    # Generate questions
+                    questions = self.question_generator.generate_questions(
+                        current_knowledge=context,
+                        query=query,
+                        questions_per_iteration=int(
+                            get_db_setting("search.questions_per_iteration")
+                        ),
+                        questions_by_iteration=self.questions_by_iteration,
+                    )
 
-            # Function for thread pool
-            def search_question(q):
-                try:
-                    result = self.search.run(q)
-                    return {"question": q, "results": result or []}
-                except Exception as e:
-                    logger.error(f"Error searching for '{q}': {str(e)}")
-                    return {"question": q, "results": [], "error": str(e)}
-
-            # Run searches in parallel
-            with concurrent.futures.ThreadPoolExecutor(
-                max_workers=len(all_questions)
-            ) as executor:
-                futures = [executor.submit(search_question, q) for q in all_questions]
-                all_search_dict = {}
-
-                # Process results as they complete
-                for i, future in enumerate(concurrent.futures.as_completed(futures)):
-                    result_dict = future.result()
-                    question = result_dict["question"]
-                    search_results = result_dict["results"]
-                    all_search_dict[question] = search_results
+                    # Use only the new questions for this iteration's searches
+                    all_questions = questions
 
-                    self._update_progress(
-                        f"Completed search {i + 1} of {len(all_questions)}: {question[:30]}...",
-                        20 + ((i + 1) / len(all_questions) * 40),
-                        {
-                            "phase": "search_complete",
-                            "result_count": len(search_results),
-                            "question": question,
-                        },
+                    # Store in questions_by_iteration
+                    self.questions_by_iteration[iteration] = questions
+                    logger.info(
+                        f"Generated questions for iteration {iteration}: {questions}"
                     )
 
-                    # Extract and save links
-                    if not self.use_cross_engine_filter:
-                        links = extract_links_from_search_results(search_results)
-                        self.all_links_of_system.extend(links)
-                    all_search_results.extend(search_results)
+                # Step 2: Run all searches in parallel for this iteration
+                self._update_progress(
+                    f"Running parallel searches for iteration {iteration}",
+                    iteration_progress_base + 10,
+                    {"phase": "parallel_search", "iteration": iteration},
+                )
 
-            # Step 3: Analysis of collected search results
-            self._update_progress(
-                "Analyzing all collected search results",
-                70,
-                {"phase": "final_analysis"},
-            )
-            if self.use_cross_engine_filter:
+                # Function for thread pool
+                def search_question(q):
+                    try:
+                        result = self.search.run(q)
+                        return {"question": q, "results": result or []}
+                    except Exception as e:
+                        logger.error(f"Error searching for '{q}': {str(e)}")
+                        return {"question": q, "results": [], "error": str(e)}
+
+                # Run searches in parallel
+                with concurrent.futures.ThreadPoolExecutor(
+                    max_workers=len(all_questions)
+                ) as executor:
+                    futures = [
+                        executor.submit(search_question, q) for q in all_questions
+                    ]
+                    iteration_search_dict = {}
+                    iteration_search_results = []
+
+                    # Process results as they complete
+                    for i, future in enumerate(
+                        concurrent.futures.as_completed(futures)
+                    ):
+                        result_dict = future.result()
+                        question = result_dict["question"]
+                        search_results = result_dict["results"]
+                        iteration_search_dict[question] = search_results
+
+                        self._update_progress(
+                            f"Completed search {i + 1} of {len(all_questions)}: {question[:30]}...",
+                            iteration_progress_base
+                            + 10
+                            + ((i + 1) / len(all_questions) * 30),
+                            {
+                                "phase": "search_complete",
+                                "iteration": iteration,
+                                "result_count": len(search_results),
+                                "question": question,
+                            },
+                        )
+
+                        # Collect all search results for this iteration
+                        iteration_search_results.extend(search_results)
+
+                # Step 3: Filter and analyze results for this iteration
                 self._update_progress(
-                    "Filtering search results across engines",
-                    65,
-                    {"phase": "cross_engine_filtering"},
+                    f"Analyzing results for iteration {iteration}",
+                    iteration_progress_base + 45,
+                    {"phase": "iteration_analysis", "iteration": iteration},
                 )
 
-                # Get the current link count (for indexing)
-                existing_link_count = len(self.all_links_of_system)
+                # Apply cross-engine filtering if enabled
+                if self.use_cross_engine_filter:
+                    self._update_progress(
+                        f"Filtering search results for iteration {iteration}",
+                        iteration_progress_base + 45,
+                        {"phase": "cross_engine_filtering", "iteration": iteration},
+                    )
+
+                    # Get the current link count (for indexing)
+                    existing_link_count = len(self.all_links_of_system)
 
-                # Filter the search results
-                filtered_search_results = self.cross_engine_filter.filter_results(
-                    all_search_results,
-                    query,
-                    reorder=self.filter_reorder,
-                    reindex=self.filter_reindex,
-                    start_index=existing_link_count,  # Start indexing after existing links
-                )
+                    # Filter the search results
+                    filtered_search_results = self.cross_engine_filter.filter_results(
+                        iteration_search_results,
+                        query,
+                        reorder=self.filter_reorder,
+                        reindex=self.filter_reindex,
+                        start_index=existing_link_count,  # Start indexing after existing links
+                    )
 
-                links = extract_links_from_search_results(filtered_search_results)
-                self.all_links_of_system.extend(links)
+                    links = extract_links_from_search_results(filtered_search_results)
+                    self.all_links_of_system.extend(links)
+
+                    self._update_progress(
+                        f"Filtered from {len(iteration_search_results)} to {len(filtered_search_results)} results",
+                        iteration_progress_base + 50,
+                        {
+                            "phase": "filtering_complete",
+                            "iteration": iteration,
+                            "links_count": len(self.all_links_of_system),
+                        },
+                    )
 
+                    # Use filtered results for analysis
+                    iteration_search_results = filtered_search_results
+                else:
+                    # Just extract links without filtering
+                    links = extract_links_from_search_results(iteration_search_results)
+                    self.all_links_of_system.extend(links)
+
+                # Add to all search results
+                all_search_results.extend(iteration_search_results)
+
+                # Create a finding for this iteration's results
+                if self.include_text_content and iteration_search_results:
+                    # For iteration > 1 with knowledge accumulation, use follow-up analysis
+                    if iteration > 1 and iterations_to_run > 2:
+                        citation_result = self.citation_handler.analyze_followup(
+                            query,
+                            iteration_search_results,
+                            current_knowledge,
+                            len(self.all_links_of_system) - len(links),
+                        )
+                    else:
+                        # For first iteration or without knowledge accumulation, use initial analysis
+                        citation_result = self.citation_handler.analyze_initial(
+                            query, iteration_search_results
+                        )
+
+                    if citation_result:
+                        # Create a finding for this iteration
+                        iteration_content = citation_result["content"]
+
+                        # Update current knowledge if iterations > 2
+                        if iterations_to_run > 2:
+                            if current_knowledge:
+                                current_knowledge = f"{current_knowledge}\n\n## FINDINGS FROM ITERATION {iteration}:\n\n{iteration_content}"
+                            else:
+                                current_knowledge = iteration_content
+
+                        finding = {
+                            "phase": f"Iteration {iteration}",
+                            "content": iteration_content,
+                            "question": query,
+                            "search_results": iteration_search_results,
+                            "documents": citation_result.get("documents", []),
+                        }
+                        findings.append(finding)
+
+                        # Add documents to repository
+                        if "documents" in citation_result:
+                            self.findings_repository.add_documents(
+                                citation_result["documents"]
+                            )
+
+                # Mark iteration as complete
+                iteration_progress = 5 + iteration * (70 / iterations_to_run)
                 self._update_progress(
-                    f"Filtered from {len(all_search_results)} to {len(filtered_search_results)} results",
-                    70,
-                    {
-                        "phase": "filtering_complete",
-                        "links_count": len(self.all_links_of_system),
-                    },
+                    f"Completed iteration {iteration}/{iterations_to_run}",
+                    iteration_progress,
+                    {"phase": "iteration_complete", "iteration": iteration},
                 )
 
-                # Use filtered results for analysis
-                all_search_results = filtered_search_results
+            # Final synthesis after all iterations
+            self._update_progress(
+                "Generating final synthesis", 80, {"phase": "synthesis"}
+            )
 
-            # Now when we use the citation handler, ensure we're using all_search_results:
+            # Handle final synthesis based on include_text_content flag
             if self.include_text_content:
-                # Use citation handler for analysis of all results together
-                citation_result = self.citation_handler.analyze_initial(
-                    query, all_search_results
-                )
-
-            if citation_result:
-                synthesized_content = citation_result["content"]
-                finding = {
+                # Generate a final synthesis from all search results
+                if iterations_to_run > 1:
+                    final_citation_result = self.citation_handler.analyze_initial(
+                        query, all_search_results
+                    )
+                    # Add null check for final_citation_result
+                    if final_citation_result:
+                        synthesized_content = final_citation_result["content"]
+                    else:
+                        synthesized_content = (
+                            "No relevant results found in final synthesis."
+                        )
+                else:
+                    # For single iteration, use the content from findings
+                    synthesized_content = (
+                        findings[0]["content"]
+                        if findings
+                        else "No relevant results found."
+                    )
+                # Add a final synthesis finding
+                final_finding = {
                     "phase": "Final synthesis",
                     "content": synthesized_content,
                     "question": query,
                     "search_results": all_search_results,
-                    "documents": citation_result.get("documents", []),
+                    "documents": [],
                 }
-                findings.append(finding)
-
-                # Transfer questions to repository
-                self.findings_repository.set_questions_by_iteration(
-                    self.questions_by_iteration
-                )
-
-                # Format findings
-                formatted_findings = self.findings_repository.format_findings_to_text(
-                    findings, synthesized_content
-                )
-
-                # Add documents to repository
-                if "documents" in citation_result:
-                    self.findings_repository.add_documents(citation_result["documents"])
-                else:
-                    synthesized_content = "No relevant results found."
-                    formatted_findings = synthesized_content
-                    finding = {
-                        "phase": "Error",
-                        "content": "No relevant results found.",
-                        "question": query,
-                        "search_results": all_search_results,
-                        "documents": [],
-                    }
-                    findings.append(finding)
+                findings.append(final_finding)
             else:
                 # Skip LLM analysis, just format the raw search results
                 synthesized_content = "LLM analysis skipped"
-                finding = {
+                final_finding = {
                     "phase": "Raw search results",
                     "content": "LLM analysis was skipped. Displaying raw search results with links.",
                     "question": query,
                     "search_results": all_search_results,
                     "documents": [],
                 }
-                findings.append(finding)
+                findings.append(final_finding)
 
-                # Transfer questions to repository
-                self.findings_repository.set_questions_by_iteration(
-                    self.questions_by_iteration
-                )
+            # Transfer questions to repository
+            self.findings_repository.set_questions_by_iteration(
+                self.questions_by_iteration
+            )
 
-                # Format findings without synthesis
-                formatted_findings = self.findings_repository.format_findings_to_text(
-                    findings, "Raw search results (LLM analysis skipped)"
-                )
+            # Format findings
+            formatted_findings = self.findings_repository.format_findings_to_text(
+                findings, synthesized_content
+            )
 
         except Exception as e:
             import traceback
@@ -305,8 +434,8 @@ class ParallelSearchStrategy(BaseSearchStrategy):
 
         return {
             "findings": findings,
-            "iterations": 1,
-            "questions": self.questions_by_iteration,
+            "iterations": iterations_to_run,
+            "questions_by_iteration": self.questions_by_iteration,
             "formatted_findings": formatted_findings,
             "current_knowledge": synthesized_content,
         }

local_deep_research/advanced_search_system/strategies/rapid_search_strategy.py

@@ -23,13 +23,14 @@ class RapidSearchStrategy(BaseSearchStrategy):
     a single synthesis step at the end, optimized for speed.
     """
 
-    def __init__(self, search=None, model=None, citation_handler=None):
+    def __init__(
+        self, search=None, model=None, citation_handler=None, all_links_of_system=None
+    ):
         """Initialize with optional dependency injection for testing."""
-        super().__init__()
+        super().__init__(all_links_of_system=all_links_of_system)
         self.search = search or get_search()
         self.model = model or get_llm()
         self.progress_callback = None
-        self.all_links_of_system = list()
         self.questions_by_iteration = {}
 
         # Use provided citation_handler or create one