Creates a general-purpose chat model with no tools.

Create a Configuration instance from a RunnableConfig.

Internal helper to apply retry and runtime configuration.

Create and return the biographic event detection graph.

Creates a model configured specifically for tool-calling.

Specifically converts data to CategoriesWithEvents model.

Extract domain from URL.

Splits text into token-based, overlapping chunks.

Creates a small model for chunk biographical event detection.

Converts a dictionary to a Pydantic model instance if needed. If the data is already an instance of the model class, returns it as-is. Args:

Get API key for a specific model from environment or config.

Return a readable transcript showing roles, including tool names for ToolMessages.

Get the LLM chunk model, using overrides if provided.

Get the LLM structured model, using overrides if provided.

Get the LLM with tools model, using overrides if provided.

Merge multiple categorized event results into one.

Provide a reusable mock for merge_events_app with configurable responses.

Provide a reusable mock for model_for_structured with configurable responses.

Provide a reusable mock for url_crawler_app with configurable responses.

Removes Markdown links, keeping only display text.

Scrapes URL using Firecrawl API and returns Markdown content.

Remove first URL from list and track its domain.

Crawls a URL and returns its content. For this example, returns dummy text.

Make the mock tool call subscriptable.

Initialize mock tool call with name and args.

Initialize mock tool response with tool calls.

Initialize mock response with content.

Initialize mock response with content.

Initialize mock tool call with name and args.

Initialize mock tool response with tool calls.

Check each chunk for biographical events using structured output.

Merge original and new events for each category using an LLM.

Counts the total tokens in a list of messages.

Crawls the next URL and updates the temporary state with new events.

Create a configured mock crawler.

Create a configured mock merger.

Create a configured mock model.

Combined extraction and categorization

Filter chunks to only process those containing biographical events

Get the tiktoken tokenizer, loading it lazily.

Merge all categorized chunks into a single CategoriesWithEvents

Merges new events, removes the processed URL, and loops back to the router.

Provide mock scraped content for testing.

Reducer function that allows a new value to completely replace the old one.

Provide a sample input state for the url_crawler_app graph.

Provide a sample input state for the merge_events_app graph.

Provide a sample input state for the research_events_app graph.

Provide a sample input state for the enhanced merge events graph.

Scrapes URL content and returns it without any processing.

Use token-based chunking from URL crawler and filter for biographical events

Split events text into chunks of specified length.

Split text into smaller chunks.

Step 2: Structures the cleaned events into JSON format. Args: state: Current researcher state with cleaned events text. config: R

The 'brain' of the agent. It decides the next action.

The 'hands' of the agent. Executes tools and returns a Command for routing.

Test enhanced merge events with empty extracted content.

Unit test for the enhanced merge events graph with mocked dependencies.

Unit test for the merge events graph with a mocked LLM.

Integration test for the merge events graph with real LLM calls.

Unit test for the research events graph with mocked dependencies.

Integration test for the research events graph with real LLM calls.

Test URL crawler with empty scraped content.

Test URL crawler with content longer than MAX_CONTENT_LENGTH.

Unit test for the simplified URL crawler graph.

Test URL crawler with mocked URL crawling.

Mandatory reflection step. Use this to analyze the last result, identify gaps, and formulate the EXACT query for the next search. You MUST use th

Find the urls for the research_question