Combine all content into a single text block

Get entity metadata for entities in a community # Arguments `entity_names` - List of entity names # Returns Vec of EntityMetadata for the entities

Hash-based TF embedding implementation (Same as CandleEmbedder::embed() in graphrag-wasm/src/embedder.rs)

Generate hash-based TF embedding (same as graphrag-wasm)

Index a knowledge graph for retrieval

Compute token-based Jaccard similarity

Load knowledge graph from Parquet files

Match a query to ranked concepts Returns concepts sorted by relevance to the query, combining: - Original ranking score - Query-concept match score

Set the maximum capacity (number of entries)

Get the number of nodes in the graph

Get node degree in snapshot

Normalize string for comparison

Record a cache hit with time saved

Record query metric # Arguments `query_type` - Type of query (e.g., "semantic", "keyword", "hybrid") `latency_ms` - Query latency in milliseconds

Remove a value from the cache

Retrieve relevant communities at a specific hierarchical level # Arguments `query` - User query string `graph` - The original graph `level` - Hierarc

Save knowledge graph to Parquet files

Perform iterative deepening search Returns search results with relevant chunks and search statistics

Search for documents matching the query

Serialize the index for storage (returns JSON)

Set configuration from JSON

Get the number of indexed embeddings (if available)

Get collection statistics

Suggest hierarchical level for query

Generate a summary of the given text # Arguments `text` - The input text to summarize `max_length` - Maximum character length of the summary # Retur

Convert TOML config to EmbeddingConfig

Convert to the existing Config format for compatibility

Validate imported data

Apply strategy weight to results

Set embedding vector for this relationship (Phase 2.2)

Add metadata to the cache key

Configure for Ollama LLM (enables semantic extraction) This transitions the builder to the `HasLlm` state. Sets up Ollama with default localhost:1143

Enable Ollama with sensible defaults (localhost:11434, llama3.2:3b).

Set the top-k results for retrieval (optional)

Get workspace directory

Acquires a permit for making an LLM API call Blocks until both concurrency and rate limits allow the call to proceed. The permit must be held for the

Add a cluster to this level

Add example entity

Build the warming configuration without validation

Build the evaluable query result

Build the processor

Build the index from all added vectors

Create a registry builder from this configuration This method creates service instances based on the configuration and available features. Services a

Build report from current graph state

Get cache health status

Calculate coherence from sentence embeddings Uses a combination of: 1. Adjacent sentence similarity (weighted higher) 2. All-pairs average similarity

Calculate cohesion score for a cluster Uses internal edge density to measure how tightly connected relationships within the cluster are. Cohesion =

Calculate inverse document frequency (IDF) IDF = log(total_documents / documents_containing_term) If term is not in corpus, uses a default IDF (assu

Calculate relevance score for an anchor Combines entity count, PageRank scores, and semantic similarity

Get the total number of unique chunks

Clean and normalize text

Clear all pending updates and reset state

Combine multiple scoring signals with configurable weights

Complete a prompt with caching support (synchronous version)

Marks an operation as complete with results

Sequential similarity computation (fallback)

Convert raw configuration to the main Config struct

Cosine similarity implementation (Exact same algorithm as graphrag-wasm/src/lib.rs:50-64)

Count matching keywords between chunk and query

Create a new change record

Auto-detect and create the best available embedder Priority: 1. Burn WebGPU (if available and feature enabled) 2. Candle CPU (fallback) # Arguments

Sequential leaf node creation (fallback)

Helper function to create test vectors

Get the depth of the section number (number of components)

Detect hierarchical communities in graph

Extract K-hop ego-network around an entity An ego-network is a subgraph containing all entities within K hops of the source entity. This is useful fo

Get reference to entities (for internal use)

Get the total number of unique entities

Evaluate query quality using heuristic metrics Provides fast, deterministic evaluation without requiring LLM calls.

Set the eviction policy