local-deep-research 0.5.9__py3-none-any.whl → 0.6.0__py3-none-any.whl

local_deep_research/config/llm_config.py

@@ -3,7 +3,7 @@ from functools import cache
 
 from langchain_anthropic import ChatAnthropic
 from langchain_community.llms import VLLM
-from langchain_core.language_models import FakeListChatModel
+from langchain_core.language_models import FakeListChatModel, BaseChatModel
 from langchain_ollama import ChatOllama
 from langchain_openai import ChatOpenAI
 from loguru import logger
@@ -11,6 +11,7 @@ from loguru import logger
 from ..utilities.db_utils import get_db_setting
 from ..utilities.search_utilities import remove_think_tags
 from ..utilities.url_utils import normalize_url
+from ..llm import get_llm_from_registry, is_llm_registered
 
 # Valid provider options
 VALID_PROVIDERS = [
@@ -206,16 +207,6 @@ def get_llm(
     if provider is None:
         provider = get_db_setting("llm.provider", "ollama")
 
-    # Check if we're in testing mode and should use fallback
-    if os.environ.get("LDR_USE_FALLBACK_LLM", ""):
-        logger.info("LDR_USE_FALLBACK_LLM is set, using fallback model")
-        return wrap_llm_without_think_tags(
-            get_fallback_model(temperature),
-            research_id=research_id,
-            provider="fallback",
-            research_context=research_context,
-        )
-
     # Clean model name: remove quotes and extra whitespace
     if model_name:
         model_name = model_name.strip().strip("\"'").strip()
@@ -227,6 +218,74 @@ def get_llm(
     # Normalize provider: convert to lowercase
     provider = provider.lower() if provider else None
 
+    # Check if this is a registered custom LLM first
+    if provider and is_llm_registered(provider):
+        logger.info(f"Using registered custom LLM: {provider}")
+        custom_llm = get_llm_from_registry(provider)
+
+        # Check if it's already a BaseChatModel instance
+        if isinstance(custom_llm, BaseChatModel):
+            # It's already an LLM instance, use it directly
+            llm_instance = custom_llm
+        elif callable(custom_llm):
+            # It's a factory function, call it with parameters
+            try:
+                llm_instance = custom_llm(
+                    model_name=model_name,
+                    temperature=temperature,
+                )
+            except Exception as e:
+                logger.exception(f"Error creating custom LLM instance: {e}")
+                raise
+        else:
+            raise ValueError(
+                f"Registered LLM {provider} is neither a BaseChatModel nor a callable factory"
+            )
+
+        return wrap_llm_without_think_tags(
+            llm_instance,
+            research_id=research_id,
+            provider=provider,
+            research_context=research_context,
+        )
+
+    # Check if we're in testing mode and should use fallback (but only when no API keys are configured)
+    if os.environ.get("LDR_USE_FALLBACK_LLM", ""):
+        # Only use fallback if the provider has no valid configuration
+        provider_has_config = False
+
+        if provider == "openai" and get_db_setting("llm.openai.api_key"):
+            provider_has_config = True
+        elif provider == "anthropic" and get_db_setting(
+            "llm.anthropic.api_key"
+        ):
+            provider_has_config = True
+        elif provider == "openai_endpoint" and get_db_setting(
+            "llm.openai_endpoint.api_key"
+        ):
+            provider_has_config = True
+        elif provider == "ollama" and is_ollama_available():
+            provider_has_config = True
+        elif provider in ["vllm", "lmstudio", "llamacpp"]:
+            # These are local providers, check their availability
+            if provider == "vllm" and is_vllm_available():
+                provider_has_config = True
+            elif provider == "lmstudio" and is_lmstudio_available():
+                provider_has_config = True
+            elif provider == "llamacpp" and is_llamacpp_available():
+                provider_has_config = True
+
+        if not provider_has_config:
+            logger.info(
+                "LDR_USE_FALLBACK_LLM is set and no valid provider config found, using fallback model"
+            )
+            return wrap_llm_without_think_tags(
+                get_fallback_model(temperature),
+                research_id=research_id,
+                provider="fallback",
+                research_context=research_context,
+            )
+
     # Validate provider
     if provider not in VALID_PROVIDERS:
         logger.error(f"Invalid provider in settings: {provider}")
@@ -242,16 +301,37 @@ def get_llm(
         "temperature": temperature,
     }
 
-    # Get context window size from settings
-    context_window_size = get_db_setting("llm.context_window_size", 32000)
+    # Get context window size from settings (use different defaults for local vs cloud providers)
+    def get_context_window_size(provider_type):
+        if provider_type in ["ollama", "llamacpp", "lmstudio"]:
+            # Local providers: use smaller default to prevent memory issues
+            return get_db_setting("llm.local_context_window_size", 4096)
+        else:
+            # Cloud providers: check if unrestricted mode is enabled
+            use_unrestricted = get_db_setting(
+                "llm.context_window_unrestricted", True
+            )
+            if use_unrestricted:
+                # Let cloud providers auto-handle context (return None or very large value)
+                return None  # Will be handled per provider
+            else:
+                # Use user-specified limit
+                return get_db_setting("llm.context_window_size", 128000)
+
+    context_window_size = get_context_window_size(provider)
 
     if get_db_setting("llm.supports_max_tokens", True):
         # Use 80% of context window to leave room for prompts
-        max_tokens = min(
-            int(get_db_setting("llm.max_tokens", 30000)),
-            int(context_window_size * 0.8),
-        )
-        common_params["max_tokens"] = max_tokens
+        if context_window_size is not None:
+            max_tokens = min(
+                int(get_db_setting("llm.max_tokens", 100000)),
+                int(context_window_size * 0.8),
+            )
+            common_params["max_tokens"] = max_tokens
+        else:
+            # Unrestricted context: use provider's default max_tokens
+            max_tokens = int(get_db_setting("llm.max_tokens", 100000))
+            common_params["max_tokens"] = max_tokens
 
     # Handle different providers
     if provider == "anthropic":
@@ -301,6 +381,7 @@ def get_llm(
             openai_endpoint_url = get_db_setting(
                 "llm.openai_endpoint.url", "https://openrouter.ai/api/v1"
             )
+        openai_endpoint_url = normalize_url(openai_endpoint_url)
 
         llm = ChatOpenAI(
             model=model_name,
@@ -394,8 +475,12 @@ def get_llm(
                 f"Creating ChatOllama with model={model_name}, base_url={base_url}"
             )
             try:
+                # Add num_ctx parameter for Ollama context window size
+                ollama_params = {**common_params}
+                if context_window_size is not None:
+                    ollama_params["num_ctx"] = context_window_size
                 llm = ChatOllama(
-                    model=model_name, base_url=base_url, **common_params
+                    model=model_name, base_url=base_url, **ollama_params
                 )
 
                 # Log the actual client configuration after creation
@@ -475,7 +560,7 @@ def get_llm(
             llm = LlamaCppClient(
                 server_url=server_url,
                 temperature=temperature,
-                max_tokens=get_db_setting("llm.max_tokens", 30000),
+                max_tokens=get_db_setting("llm.max_tokens", 8192),
             )
         else:
             # Use direct model loading (existing code)
@@ -498,7 +583,7 @@ def get_llm(
                 n_gpu_layers=n_gpu_layers,
                 n_batch=n_batch,
                 f16_kv=f16_kv,
-                n_ctx=context_window_size,  # Set context window size directly
+                n_ctx=context_window_size,  # Set context window size directly (None = use default)
                 verbose=True,
             )
 

local_deep_research/defaults/default_settings.json

@@ -194,13 +194,27 @@
         "value": 30000,
         "visible": true
     },
+    "llm.context_window_unrestricted": {
+        "category": "llm_parameters",
+        "description": "Let cloud providers automatically handle context sizing (recommended). Uncheck to set a specific limit.",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Use Unrestricted Cloud Context Window",
+        "options": null,
+        "step": null,
+        "type": "LLM",
+        "ui_element": "checkbox",
+        "value": true,
+        "visible": true
+    },
     "llm.context_window_size": {
         "category": "llm_parameters",
-        "description": "Maximum context window size in tokens for the LLM",
+        "description": "Maximum context window size in tokens for cloud LLMs. Only used when unrestricted context is disabled.",
         "editable": true,
         "max_value": 20000000.0,
         "min_value": 512.0,
-        "name": "Context Window Size",
+        "name": "Cloud Provider Context Window Limit",
         "options": null,
         "step": null,
         "type": "LLM",
@@ -208,6 +222,20 @@
         "value": 128000,
         "visible": true
     },
+    "llm.local_context_window_size": {
+        "category": "llm_parameters",
+        "description": "Context window size in tokens for local LLMs (Ollama, LlamaCpp). Smaller values prevent memory issues.",
+        "editable": true,
+        "max_value": 131072.0,
+        "min_value": 512.0,
+        "name": "Local Provider Context Window Size",
+        "options": null,
+        "step": 512.0,
+        "type": "LLM",
+        "ui_element": "number",
+        "value": 4096,
+        "visible": true
+    },
     "llm.supports_max_tokens": {
         "category": "llm_parameters",
         "description": "Whether the LLM API supports the 'max_tokens' option.",
@@ -4037,6 +4065,197 @@
         "value": false,
         "visible": true
     },
+    "search.engine.web.tavily.display_name": {
+        "category": "tavily",
+        "description": "Display name to use in the U.I. for this search engine.",
+        "editable": false,
+        "max_value": null,
+        "min_value": null,
+        "name": "Display Name",
+        "options": null,
+        "step": null,
+        "type": "SEARCH",
+        "ui_element": "text",
+        "value": "Tavily",
+        "visible": false
+    },
+    "search.engine.web.tavily.description": {
+        "category": "tavily",
+        "description": "Human-readable description of the search engine.",
+        "editable": false,
+        "max_value": null,
+        "min_value": null,
+        "name": "Description",
+        "options": null,
+        "step": null,
+        "type": "SEARCH",
+        "ui_element": "text",
+        "value": "AI-powered search engine optimized for research with built-in answer extraction.",
+        "visible": false
+    },
+    "search.engine.web.tavily.api_key": {
+        "category": "tavily",
+        "description": "The Tavily API key to use.",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Api Key",
+        "options": null,
+        "step": null,
+        "type": "SEARCH",
+        "ui_element": "password",
+        "value": "TAVILY_API_KEY",
+        "visible": true
+    },
+    "search.engine.web.tavily.class_name": {
+        "category": "tavily",
+        "description": "Setting for tavily.class_name",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Class Name",
+        "options": null,
+        "step": null,
+        "type": "SEARCH",
+        "ui_element": "text",
+        "value": "TavilySearchEngine",
+        "visible": true
+    },
+    "search.engine.web.tavily.default_params.search_depth": {
+        "category": "tavily",
+        "description": "Search depth - basic for speed, advanced for quality",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Search Depth",
+        "options": ["basic", "advanced"],
+        "step": null,
+        "type": "SEARCH",
+        "ui_element": "select",
+        "value": "basic",
+        "visible": true
+    },
+    "search.engine.web.tavily.default_params.include_full_content": {
+        "category": "tavily",
+        "description": "Include full webpage content in results",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Include Full Content",
+        "options": null,
+        "step": null,
+        "type": "SEARCH",
+        "ui_element": "checkbox",
+        "value": true,
+        "visible": true
+    },
+    "search.engine.web.tavily.module_path": {
+        "category": "tavily",
+        "description": "Setting for tavily.module_path",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Module Path",
+        "options": null,
+        "step": null,
+        "type": "SEARCH",
+        "ui_element": "text",
+        "value": ".engines.search_engine_tavily",
+        "visible": true
+    },
+    "search.engine.web.tavily.requires_api_key": {
+        "category": "tavily",
+        "description": "Setting for tavily.requires_api_key",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Requires Api Key",
+        "options": null,
+        "step": null,
+        "type": "SEARCH",
+        "ui_element": "checkbox",
+        "value": true,
+        "visible": true
+    },
+    "search.engine.web.tavily.reliability": {
+        "category": "tavily",
+        "description": "Setting for tavily.reliability",
+        "editable": true,
+        "max_value": 1.0,
+        "min_value": 0.0,
+        "name": "Reliability",
+        "options": null,
+        "step": 0.05,
+        "type": "SEARCH",
+        "ui_element": "range",
+        "value": 0.8,
+        "visible": true
+    },
+    "search.engine.web.tavily.strengths": {
+        "category": "tavily",
+        "description": "Setting for tavily.strengths",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Strengths",
+        "options": null,
+        "step": null,
+        "type": "SEARCH",
+        "ui_element": "json",
+        "value": [
+            "AI-powered search optimization",
+            "built-in answer extraction",
+            "research-focused results",
+            "high-quality content filtering",
+            "fast response times"
+        ],
+        "visible": true
+    },
+    "search.engine.web.tavily.supports_full_search": {
+        "category": "tavily",
+        "description": "Setting for tavily.supports_full_search",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Supports Full Search",
+        "options": null,
+        "step": null,
+        "type": "SEARCH",
+        "ui_element": "checkbox",
+        "value": true,
+        "visible": true
+    },
+    "search.engine.web.tavily.weaknesses": {
+        "category": "tavily",
+        "description": "Setting for tavily.weaknesses",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Weaknesses",
+        "options": null,
+        "step": null,
+        "type": "SEARCH",
+        "ui_element": "json",
+        "value": [
+            "requires API key with usage limits",
+            "newer service with smaller historical data"
+        ],
+        "visible": true
+    },
+    "search.engine.web.tavily.use_in_auto_search": {
+        "category": "tavily",
+        "description": "Include Tavily in auto search mode",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Include in Auto Search",
+        "options": null,
+        "step": null,
+        "type": "SEARCH",
+        "ui_element": "checkbox",
+        "value": false,
+        "visible": true
+    },
     "search.engine.local.local_all.use_in_auto_search": {
         "category": "local_all",
         "description": "Include local documents in auto search mode",
@@ -4129,5 +4348,231 @@
         "ui_element": "text",
         "value": "http://localhost:8000",
         "visible": true
+    },
+    "app.warnings.dismiss_high_context": {
+        "category": "warnings",
+        "description": "Dismiss warnings about high context window sizes that may cause memory issues",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Dismiss High Context Warnings",
+        "options": null,
+        "step": null,
+        "type": "APP",
+        "ui_element": "checkbox",
+        "value": false,
+        "visible": true
+    },
+    "app.warnings.dismiss_low_context_focused": {
+        "category": "warnings",
+        "description": "Dismiss warnings about using focused iteration with low context window sizes",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Dismiss Low Context + Focused Strategy Warnings",
+        "options": null,
+        "step": null,
+        "type": "APP",
+        "ui_element": "checkbox",
+        "value": false,
+        "visible": true
+    },
+    "app.warnings.dismiss_model_mismatch": {
+        "category": "warnings",
+        "description": "Dismiss warnings about context size vs model size mismatches",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Dismiss Model Mismatch Warnings",
+        "options": null,
+        "step": null,
+        "type": "APP",
+        "ui_element": "checkbox",
+        "value": false,
+        "visible": true
+    },
+    "app.warnings.dismiss_searxng_recommendation": {
+        "category": "warnings",
+        "description": "Dismiss recommendations about using more questions instead of iterations with SearXNG",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Dismiss SearXNG Optimization Tips",
+        "options": null,
+        "step": null,
+        "type": "APP",
+        "ui_element": "checkbox",
+        "value": false,
+        "visible": true
+    },
+    "rate_limiting.enabled": {
+        "category": "rate_limiting",
+        "description": "Enable adaptive rate limiting system that learns optimal wait times for each search engine",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Enable Rate Limiting",
+        "options": null,
+        "step": null,
+        "type": "APP",
+        "ui_element": "checkbox",
+        "value": true,
+        "visible": true
+    },
+    "rate_limiting.exploration_rate": {
+        "category": "rate_limiting",
+        "description": "Percentage of attempts that will explore faster rates to discover improvements (0.0-1.0)",
+        "editable": true,
+        "max_value": 1.0,
+        "min_value": 0.0,
+        "name": "Exploration Rate",
+        "options": null,
+        "step": 0.05,
+        "type": "APP",
+        "ui_element": "range",
+        "value": 0.1,
+        "visible": true
+    },
+    "rate_limiting.learning_rate": {
+        "category": "rate_limiting",
+        "description": "How quickly to adapt to new rate limit information (higher = faster adaptation)",
+        "editable": true,
+        "max_value": 1.0,
+        "min_value": 0.05,
+        "name": "Learning Rate",
+        "options": null,
+        "step": 0.05,
+        "type": "APP",
+        "ui_element": "range",
+        "value": 0.3,
+        "visible": true
+    },
+    "rate_limiting.memory_window": {
+        "category": "rate_limiting",
+        "description": "Number of recent attempts to keep in memory for learning",
+        "editable": true,
+        "max_value": 1000,
+        "min_value": 10,
+        "name": "Memory Window",
+        "options": null,
+        "step": 10,
+        "type": "APP",
+        "ui_element": "number",
+        "value": 100,
+        "visible": true
+    },
+    "rate_limiting.profile": {
+        "category": "rate_limiting",
+        "description": "Rate limiting aggressiveness profile",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Rate Limiting Profile",
+        "options": [
+            {
+                "label": "Conservative (slower, more reliable)",
+                "value": "conservative"
+            },
+            {
+                "label": "Balanced (default)",
+                "value": "balanced"
+            },
+            {
+                "label": "Aggressive (faster, higher risk)",
+                "value": "aggressive"
+            }
+        ],
+        "step": null,
+        "type": "APP",
+        "ui_element": "select",
+        "value": "balanced",
+        "visible": true
+    },
+    "rate_limiting.decay_per_day": {
+        "category": "rate_limiting",
+        "description": "Confidence decay factor per day for old rate limit estimates (0.5-0.99, lower = faster decay)",
+        "editable": true,
+        "max_value": 0.99,
+        "min_value": 0.5,
+        "name": "Decay Factor Per Day",
+        "options": null,
+        "step": 0.01,
+        "type": "APP",
+        "ui_element": "range",
+        "value": 0.95,
+        "visible": true
+    },
+    "benchmark.evaluation.provider": {
+        "category": "benchmark",
+        "description": "Provider for benchmark evaluation model",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Evaluation Provider",
+        "options": [
+            {
+                "label": "OpenRouter (Recommended)",
+                "value": "openai_endpoint"
+            },
+            {
+                "label": "OpenAI API",
+                "value": "openai"
+            },
+            {
+                "label": "Anthropic API",
+                "value": "anthropic"
+            },
+            {
+                "label": "Ollama (Local)",
+                "value": "ollama"
+            }
+        ],
+        "step": null,
+        "type": "SEARCH",
+        "ui_element": "select",
+        "value": "openai_endpoint",
+        "visible": true
+    },
+    "benchmark.evaluation.model": {
+        "category": "benchmark",
+        "description": "Model for evaluating benchmark results",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Evaluation Model",
+        "options": null,
+        "step": null,
+        "type": "SEARCH",
+        "ui_element": "text",
+        "value": "anthropic/claude-3.7-sonnet",
+        "visible": true
+    },
+    "benchmark.evaluation.endpoint_url": {
+        "category": "benchmark",
+        "description": "Endpoint URL for evaluation model (when using OpenAI-compatible APIs)",
+        "editable": true,
+        "max_value": null,
+        "min_value": null,
+        "name": "Evaluation Endpoint URL",
+        "options": null,
+        "step": null,
+        "type": "SEARCH",
+        "ui_element": "text",
+        "value": "https://openrouter.ai/api/v1",
+        "visible": true
+    },
+    "benchmark.evaluation.temperature": {
+        "category": "benchmark",
+        "description": "Temperature for evaluation (0 recommended for consistency)",
+        "editable": true,
+        "max_value": 1.0,
+        "min_value": 0.0,
+        "name": "Evaluation Temperature",
+        "options": null,
+        "step": 0.1,
+        "type": "SEARCH",
+        "ui_element": "range",
+        "value": 0,
+        "visible": true
     }
 }

local_deep_research/error_handling/report_generator.py

@@ -346,6 +346,16 @@ We're here to help you get this working:
                 "- Check search engine settings in Advanced Options\n"
                 "- Ensure required API keys are set for external search engines"
             ),
+            "No search results found|All search engines.*blocked.*rate.*limited": (
+                "No search results were found for your query. This could mean all search engines are unavailable.\n\n"
+                "**Try this:**\n"
+                "- **If using SearXNG:** Check if your SearXNG Docker container is running: `docker ps`\n"
+                "- **Start SearXNG:** `docker run -d -p 8080:8080 searxng/searxng` then set URL to `http://localhost:8080`\n"
+                "- **Try different search terms:** Use broader, more general keywords\n"
+                "- **Check network connection:** Ensure you can access the internet\n"
+                "- **Switch search engines:** Try DuckDuckGo, Brave, or Google (if API key configured)\n"
+                "- **Check for typos** in your research query"
+            ),
             "TypeError.*Context.*Size|'<' not supported between": (
                 "Model configuration issue. The context size setting might not be compatible with your model.\n\n"
                 "**Try this:**\n"