Skip to content

~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.

13.9 KB
Raw Blame
"""End-to-end tests for API settings in research workflows."""

from unittest.mock import patch, MagicMock

from local_deep_research.api import quick_summary, detailed_research
from local_deep_research.api.settings_utils import create_settings_snapshot


class TestE2EResearchWithSettings:
    """Test end-to-end research workflows with various settings."""

    @patch("local_deep_research.api.research_functions.get_llm")
    @patch("local_deep_research.api.research_functions.get_search")
    def test_quick_summary_full_flow(self, mock_get_search, mock_get_llm):
        """Test quick_summary with full settings propagation.

        Patches happen on `api.research_functions` because the names are
        imported into that module at import time — patching the original
        module (`config.llm_config.get_llm`) would not affect already-bound
        references.
        """
        # Mock LLM
        mock_llm = MagicMock()
        mock_llm.invoke.return_value = MagicMock(content="Test summary")
        mock_get_llm.return_value = mock_llm

        # Mock search engine
        mock_search = MagicMock()
        mock_search.search.return_value = {
            "results": [
                {
                    "title": "Result 1",
                    "url": "http://example.com/1",
                    "snippet": "Snippet 1",
                },
                {
                    "title": "Result 2",
                    "url": "http://example.com/2",
                    "snippet": "Snippet 2",
                },
            ]
        }
        mock_get_search.return_value = mock_search

        # Run quick summary with custom settings
        result = quick_summary(
            "What is quantum computing?",
            provider="anthropic",
            api_key="test-key",
            temperature=0.5,
            max_search_results=10,
            settings_override={
                "llm.anthropic.model": "claude-3-opus-20240229",
                "search.tool": "duckduckgo",
                "search.region": "us-en",
            },
        )

        # Verify LLM was configured correctly
        mock_get_llm.assert_called()

        # Verify search was configured correctly. get_search() takes the
        # tool name as the first positional arg; everything else is in
        # settings_snapshot rather than dedicated kwargs.
        mock_get_search.assert_called()
        search_args, search_kwargs = mock_get_search.call_args
        assert search_args[0] == "duckduckgo"
        snapshot = search_kwargs["settings_snapshot"]
        assert snapshot["search.region"]["value"] == "us-en"
        assert snapshot["search.max_results"]["value"] == 10

        # Verify result structure
        assert "summary" in result
        assert "findings" in result
        assert "iterations" in result

    @patch("local_deep_research.api.research_functions.get_llm")
    @patch("local_deep_research.api.research_functions.get_search")
    def test_detailed_research_full_flow(self, mock_get_search, mock_get_llm):
        """Test detailed_research with comprehensive settings.

        detailed_research accepts only `settings_snapshot` for configuration
        (unlike quick_summary which has provider/api_key shortcuts), so the
        test builds a snapshot via create_settings_snapshot and threads it
        in directly. Patch targets follow `api.research_functions` because
        that is where get_llm / get_search are imported and bound.
        """
        # Mock LLM with different responses
        mock_llm = MagicMock()
        mock_llm.invoke.side_effect = [
            MagicMock(content="Initial analysis"),
            MagicMock(content="Deeper analysis"),
            MagicMock(content="Final synthesis"),
        ]
        mock_get_llm.return_value = mock_llm

        mock_search = MagicMock()
        mock_get_search.return_value = mock_search

        snapshot = create_settings_snapshot(
            provider="openai",
            overrides={
                "search.iterations": 3,
                "search.max_results": 20,
                "search.engines.arxiv.enabled": True,
                "llm.max_tokens": 4000,
            },
        )

        # Run detailed research with custom settings
        result = detailed_research(
            "Explain the applications of quantum computing in cryptography",
            settings_snapshot=snapshot,
        )

        # The settings_snapshot values should be threaded through to
        # get_search.
        mock_get_search.assert_called()
        passed_snapshot = mock_get_search.call_args[1]["settings_snapshot"]
        assert passed_snapshot["search.max_results"]["value"] == 20
        assert passed_snapshot["search.iterations"]["value"] == 3

        # Verify the structured-report shape is returned (detailed_research
        # returns a report dict, not a summary dict).
        assert isinstance(result, dict)

    def test_settings_isolation_between_calls(self):
        """Test that settings don't leak between API calls."""
        with patch(
            "local_deep_research.api.research_functions._init_search_system"
        ) as mock_init:
            mock_system = MagicMock()
            mock_system.analyze_topic.return_value = {
                "current_knowledge": "Summary",
                "findings": [],
                "iterations": 1,
                "questions": {},
                "formatted_findings": "Findings",
                "all_links_of_system": [],
            }
            mock_init.return_value = mock_system

            # First call with one set of settings
            _ = quick_summary("Query 1", provider="openai", temperature=0.9)

            # Second call with different settings
            _ = quick_summary("Query 2", provider="anthropic", temperature=0.1)

            # Verify each call got its own settings
            call1_settings = mock_init.call_args_list[0][1]["settings_snapshot"]
            call2_settings = mock_init.call_args_list[1][1]["settings_snapshot"]

            assert call1_settings["llm.provider"]["value"] == "openai"
            assert call1_settings["llm.temperature"]["value"] == 0.9

            assert call2_settings["llm.provider"]["value"] == "anthropic"
            assert call2_settings["llm.temperature"]["value"] == 0.1


class TestMultiProviderScenarios:
    """Test scenarios with multiple LLM providers."""

    def test_multi_model_research(self):
        """Test research using multiple models for comparison."""
        models = [
            ("openai", "gpt-4", 0.3),
            ("anthropic", "claude-3-opus-20240229", 0.3),
            ("openai", "gpt-3.5-turbo", 0.5),
        ]

        results = []

        with patch(
            "local_deep_research.api.research_functions._init_search_system"
        ) as mock_init:
            mock_system = MagicMock()
            mock_system.analyze_topic.return_value = {
                "current_knowledge": "Model-specific summary",
                "findings": [],
                "iterations": 1,
                "questions": {},
                "formatted_findings": "Findings",
                "all_links_of_system": [],
            }
            mock_init.return_value = mock_system

            for provider, model, temp in models:
                result = quick_summary(
                    "Compare quantum and classical computing",
                    provider=provider,
                    temperature=temp,
                    settings_override={
                        f"llm.{provider}.model": model,
                    },
                )
                results.append((provider, model, result))

            # Verify each call used different settings
            assert mock_init.call_count == len(models)

            for i, (provider, model, _) in enumerate(models):
                call_settings = mock_init.call_args_list[i][1][
                    "settings_snapshot"
                ]
                assert call_settings["llm.provider"]["value"] == provider
                assert call_settings[f"llm.{provider}.model"]["value"] == model


class TestSearchEngineIntegration:
    """Test integration with various search engines through settings."""

    @patch("local_deep_research.api.research_functions.get_llm")
    @patch("local_deep_research.api.research_functions.get_search")
    def test_search_engine_specific_settings(
        self, mock_get_search, mock_get_llm
    ):
        """Test that search engine specific settings are applied.

        Each call's positional first argument identifies the chosen tool;
        engine-specific settings should appear in `settings_snapshot`.
        Patches target `api.research_functions` since that is where the
        names are bound.
        """
        search_configs = [
            {
                "engine": "searxng",
                "settings": {
                    "search.tool": "searxng",
                    "search.engines.searxng.base_url": "https://searx.example.com",
                    "search.engines.searxng.timeout": 15,
                },
            },
            {
                "engine": "duckduckgo",
                "settings": {
                    "search.tool": "duckduckgo",
                    "search.engines.duckduckgo.region": "uk-en",
                    "search.engines.duckduckgo.safe_search": False,
                },
            },
            {
                "engine": "wikipedia",
                "settings": {
                    "search.tool": "wikipedia",
                    "search.engines.wikipedia.language": "es",
                    "search.engines.wikipedia.max_chars": 2000,
                },
            },
        ]

        mock_get_llm.return_value = MagicMock()
        mock_get_search.return_value = MagicMock()

        for config in search_configs:
            _ = quick_summary(
                f"Test with {config['engine']}",
                provider="openai",
                api_key="test-key",
                settings_override=config["settings"],
            )

            # Verify the chosen search engine appears as the first positional
            # argument (the get_search public signature).
            last_args, last_kwargs = mock_get_search.call_args
            assert last_args[0] == config["engine"]

            # Engine-specific overrides must propagate through settings_snapshot.
            snapshot = last_kwargs["settings_snapshot"]
            for key, value in config["settings"].items():
                assert key in snapshot, (
                    f"missing snapshot key {key} for {config['engine']}"
                )
                assert snapshot[key]["value"] == value


class TestPerformanceSettings:
    """Test performance-related settings."""

    def test_concurrent_research_settings(self):
        """Test settings for concurrent research operations."""
        # Settings optimized for concurrent operations
        concurrent_settings = create_settings_snapshot(
            overrides={
                "research.concurrent.enabled": True,
                "research.concurrent.max_workers": 5,
                "llm.request_timeout": 30,
                "llm.max_retries": 2,
                "search.request_timeout": 10,
                "search.concurrent_searches": 3,
            }
        )

        # Verify concurrent settings
        assert (
            concurrent_settings["research.concurrent.enabled"]["value"] is True
        )
        assert (
            concurrent_settings["research.concurrent.max_workers"]["value"] == 5
        )
        assert concurrent_settings["llm.request_timeout"]["value"] == 30

    def test_rate_limiting_settings(self):
        """Test rate limiting settings for API calls."""
        rate_limit_settings = create_settings_snapshot(
            overrides={
                "llm.rate_limit.enabled": True,
                "llm.rate_limit.requests_per_minute": 60,
                "llm.rate_limit.tokens_per_minute": 90000,
                "llm.rate_limit.concurrent_requests": 5,
                "search.rate_limit.requests_per_second": 10,
            }
        )

        # Verify rate limiting configuration
        assert rate_limit_settings["llm.rate_limit.enabled"]["value"] is True
        assert (
            rate_limit_settings["llm.rate_limit.requests_per_minute"]["value"]
            == 60
        )
        assert (
            rate_limit_settings["search.rate_limit.requests_per_second"][
                "value"
            ]
            == 10
        )


class TestDebugAndMonitoringSettings:
    """Test debug and monitoring related settings."""

    def test_debug_mode_settings(self):
        """Test settings for debug mode."""
        debug_settings = create_settings_snapshot(
            overrides={
                "debug.enabled": True,
                "debug.log_level": "DEBUG",
                "debug.log_api_calls": True,
                "debug.log_search_queries": True,
                "debug.save_intermediate_results": True,
                "debug.pretty_print_responses": True,
            }
        )

        # Verify debug settings
        assert debug_settings["debug.enabled"]["value"] is True
        assert debug_settings["debug.log_level"]["value"] == "DEBUG"
        assert debug_settings["debug.log_api_calls"]["value"] is True

    def test_monitoring_settings(self):
        """Test settings for monitoring and metrics."""
        monitoring_settings = create_settings_snapshot(
            overrides={
                "monitoring.enabled": True,
                "monitoring.metrics.api_latency": True,
                "monitoring.metrics.token_usage": True,
                "monitoring.metrics.search_performance": True,
                "monitoring.export.format": "prometheus",
                "monitoring.export.endpoint": "http://metrics.example.com",
            }
        )

        # Verify monitoring configuration
        assert monitoring_settings["monitoring.enabled"]["value"] is True
        assert (
            monitoring_settings["monitoring.metrics.token_usage"]["value"]
            is True
        )
        assert (
            monitoring_settings["monitoring.export.format"]["value"]
            == "prometheus"
        )