Struct FactExtractor 

pub struct FactExtractor { /* private fields */ }

Fact extractor for knowledge graph generation.

Uses an LLM (via the Llm trait) to extract structured facts from text. Produces a KnowledgeGraph containing nodes (entities) and edges (relationships).

Example

use cognee_cognify::FactExtractor;
use cognee_llm::OpenAIAdapter;
use std::sync::Arc;

let llm = Arc::new(OpenAIAdapter::new("gpt-4", "sk-...", None)?);
let extractor = FactExtractor::new(llm);

let text = "Alice works at TechCorp in San Francisco.";
let graph = extractor.extract_facts(text, None).await?;

println!("Extracted {} nodes and {} edges", graph.node_count(), graph.edge_count());

Implementations

impl FactExtractor

pub fn new(llm: Arc<dyn Llm>) -> FactExtractor

Create a new fact extractor with the given LLM.

Arguments

• llm - An LLM implementation (e.g., OpenAIAdapter, OllamaAdapter)

Returns

A new FactExtractor instance

pub fn default_graph_prompt() -> &'static str

Return the default graph extraction prompt used by extract_facts.

pub async fn extract<M>( &self, text: &str, custom_prompt: Option<&str>, ) -> Result<M, CognifyError>

where M: GraphModel,

Extract a structured model from text via LLM.

Generic counterpart of extract_facts. Works with any type implementing GraphModel, which requires Serialize + DeserializeOwned + JsonSchema + Clone + Send + Sync.

The LLM’s create_structured_output method infers the JSON schema from M and deserializes the response into the concrete type.

No post-processing is applied; for the default KnowledgeGraph flow with name-fallback logic, use extract_facts.

Arguments

• text - Input text to extract from
• custom_prompt - Optional custom system prompt (uses [DEFAULT_GRAPH_PROMPT] if None)

Errors

Returns CognifyError::LlmError if the LLM call fails

pub async fn extract_facts( &self, text: &str, custom_prompt: Option<&str>, ) -> Result<KnowledgeGraph, CognifyError>

Extract facts (knowledge graph) from text.

Mirrors Python’s extract_content_graph function. Uses the LLM to extract structured Node and Edge objects from the input text.

Arguments

• text - Input text to extract facts from
• custom_prompt - Optional custom system prompt (uses DEFAULT_GRAPH_PROMPT if None)

Returns

A KnowledgeGraph containing extracted nodes and edges

Errors

Returns CognifyError::LlmError if the LLM call fails

pub async fn extract_facts_batch( &self, texts: Vec<String>, custom_prompt: Option<String>, ) -> Result<Vec<KnowledgeGraph>, CognifyError>

Extract facts from multiple text chunks in parallel.

Mirrors Python’s pattern in extract_graph_from_data.py where all chunks are processed concurrently using asyncio.gather.

Arguments

• texts - Slice of text strings to extract facts from
• custom_prompt - Optional custom system prompt

Returns

A vector of KnowledgeGraphs, one per input text

Errors

Returns CognifyError::LlmError if any LLM call fails

pub fn llm(&self) -> &Arc<dyn Llm>

Get a reference to the underlying LLM.

Trait Implementations

impl Clone for FactExtractor

fn clone(&self) -> FactExtractor

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.