Crate cognee_lib 

Unified public API for Cognee-Rust.

This crate provides a single entry point by re-exporting the core operations:

• add (AddPipeline)
• cognify (cognify() free function and related types)
• search (SearchBuilder/SearchOrchestrator and related types)

OpenTelemetry support

Cognee emits structured spans for every pipeline stage, search retriever, and HTTP route. To export them to an OTLP collector (Grafana Tempo, Honeycomb, Dash0, in-cluster otel-collector, …), enable the telemetry cargo feature and set OTEL_EXPORTER_OTLP_ENDPOINT:

use cognee_lib::telemetry::{init_telemetry, TelemetryGuard};
use cognee_lib::config::{ConfigManager, Settings};
use tracing_subscriber::Registry;

let settings: Settings = ConfigManager::from_env().settings().clone();
let (_layer, _guard) = init_telemetry::<Registry>(&settings)
    .expect("telemetry init");
// ... compose `_layer` onto your subscriber; spans are flushed when
// `_guard` is dropped.

See docs/observability/opentelemetry.md for the full operator guide, env-var reference, and deployment recipes.

Re-exports

pub use api::notebooks::NotebookError;

pub use api::notebooks::create_notebook;

pub use api::notebooks::delete_notebook;

pub use api::notebooks::list_notebooks;

pub use api::notebooks::update_notebook;

pub use api::DatasetDb;

pub use api::DatasetError;

pub use api::DatasetManager;

pub use component_manager::ComponentManager;

pub use config::ConfigError;

pub use config::ConfigManager;

pub use config::Settings;

pub use context::PipelineContext;

pub use error::ComponentError;

pub use cognee_core;

pub use cognee_database;

pub use cognee_delete;

pub use cognee_embedding;

pub use cognee_graph;

pub use cognee_llm;

pub use cognee_models;

pub use cognee_ontology;

pub use cognee_session;

pub use cognee_storage;

pub use cognee_vector;

pub use uuid;

Modules

add

Data ingestion (add) pipeline.

api

Top-level API functions mirroring the Python SDK. High-level API functions for Cognee-Rust.

cognify

Knowledge-graph extraction (cognify) pipeline.

component_manager

Component lifecycle management. ComponentManager: lazy-initializing, shared component store.

config

Runtime configuration management. Shared configuration types for cognee-rust.

context

Pipeline execution context. Pipeline context trait for shared component access.

core

Core pipeline orchestration primitives.

database

Metadata database abstraction.

delete

Data and dataset deletion pipeline.

embedding

Embedding engine abstraction and providers.

error

Component error types. Error types for the cognee-lib crate.

graph

Graph database abstraction.

llm

LLM provider abstraction.

models

Core data models.

ontology

Ontology resolution.

prelude

Convenience re-exports for common usage.

search

Search pipeline and retrieval strategies.

session

Session management. Public session management API — thin wrappers over SessionManager.

storage

File storage abstraction.

telemetry

Product-analytics telemetry. Telemetry surface for embedders.

vector

Vector database abstraction.

visualization

Knowledge-graph visualization (requires visualization feature).

Structs

AddParams

Optional parameters for the AddPipeline::add method.

AddPipeline

Ingest pipeline driven by the cognee-core task framework.

ContentHasher

ProcessedInput

Metadata extracted from a DataInput during streaming processing.

Enums

VisualizationError

Errors that can occur while generating an HTML graph visualization.

Functions

build_add_pipeline

Build a complete ingest Pipeline: DataInput → ProcessedInput → Data.

generate_dataset_id

Generate a deterministic dataset ID matching Python’s formula: uuid5(NAMESPACE_OID, f"{dataset_name}{str(user_id)}{str(tenant_id)}")

make_persist_data_task

Build a TypedTask that resolves or creates the dataset, deduplicates by content hash, persists a new Data record if needed, and returns it.

make_process_input_task

Build a TypedTask that streams a DataInput to storage, hashes its content, and returns a self-contained ProcessedInput.

persist_data

Persist a ProcessedInput as a Data record: resolve or create the dataset, deduplicate by content hash, create the record if new, and attach it to the dataset.

process_input

Process a single DataInput: resolve URLs, stream to storage, compute content hash, and extract all metadata needed to create a Data record.

visualize

Generate an interactive HTML knowledge-graph visualization of the supplied graph database.