Extract concepts from text

Internal entity extraction implementation

Extract chapter/section references from query

Find candidate entity for linking a new mention

Find optimal split points in text to maximize chunk coherence Uses a greedy algorithm: 1. Start with no splits 2. Try all candidate split points 3. P

Fuzzy string matching with edit distance

Generate embeddings for a batch of texts

Generate extractive answer from context with improved relevance scoring

Generate summary asynchronously

Generate warming queries based on the configured strategy

Get all chunks that contain a specific entity Returns an empty vector if the entity is not found

Get a document by ID

Get entity by ID (string version for compatibility)

Get all available function definitions

Get all vector IDs

Get predefined common queries for warming

Get statistics about lazy propagation

Get source attribution information

Get knowledge graph statistics

Get the midpoint timestamp of this step (for ordering)

Handle executing a query in Reason mode (query decomposition)

Generate hash-based TF embedding

Hash-based TF embedding implementation (Exact same algorithm as graphrag-wasm/src/embedder.rs:270-327)

Index multiple documents

Initialize intent patterns

Check if ONNX Runtime Web is available

List all workspaces

Load a document into the knowledge graph # Arguments `path` - Path to the document to load `rebuild` - If true, clears existing graph AND documents b

Build a small Document with N chunks for testing. Content is the Plato Symposium excerpt (first ~500 words) repeated across chunks.

Return true only if the user explicitly opted into Ollama tests. We use an env var instead of relying solely on `#[ignore]` so CI can never accidenta

Open the IndexedDB database. Returns `None` if the browser does not support it.

Parse LLM evaluation response

Generate a prefixed key for Redis storage

Print comprehensive cache information to the log

Process a query using the complete ROGRAG pipeline

Execute a query and return both LLM answer and raw search results Returns a tuple of (llm_answer, raw_results)

Rank all concepts in the graph Returns concepts sorted by score in descending order. # Arguments `graph` - The concept graph to rank `total_documen

Record a graph change for incremental updates

Record a cache miss

Record a query and its results

Record a successful request

Refine a query using the concept graph and bidirectional index This performs iterative deepening to expand the query with related concepts and identi

Get reference to relationships (for internal use)

Remove node and its edges

Perform hybrid search combining semantic and keyword retrieval

Serialize generation parameters for inclusion in cache key

Find shortest path between two nodes (Dijkstra's algorithm) # Arguments `start` - Starting node ID `end` - Ending node ID # Returns Shortest path if

Export statistics as a snapshot

Get snapshot at specific timestamp

Split text into sentences using multiple heuristics

Set the warming strategy to use

Submit a single update operation (non-blocking)

Synthesize multiple subquery results using the specified strategy Combines information from multiple subquery results into a single coherent response

Sequential synthesis strategy Concatenates results in order with equal weighting, computing average confidence.

Tokenize text into terms

Tokenize text into words

Set the time-to-live in seconds

Validate the configuration

Validate a query before processing

Validate relationship

Set maximum tokens for LLM generation (default: 1500)

Set minimum cluster size

Create metadata with position in document

Set resolution parameters for clustering

Register a storage implementation

Check if workspace exists

Records an access to this cache entry. Increments the access counter and updates the last_accessed timestamp.

Retrieve with detailed query analysis Returns both the retrieval results and the query analysis that determined the level. # Arguments `query` - Use

Add a custom metadata field

Register a dependency relationship

Add a document to the system

Add a document chunk with metadata # Arguments `id` - Unique document ID `embedding` - Embedding vector `metadata` - Document metadata including enti

Add a single embedding to the index # Arguments `embedding` - Vector embedding as Float32Array `id` - Document ID `title` - Document title `url` - Do