Struct ComponentHandles 

pub struct ComponentHandles {

    pub database: Arc<DatabaseConnection>,
    pub acl_db: Option<Arc<dyn AclDb>>,
    pub storage: Arc<dyn StorageTrait>,
    pub delete_service: Arc<DeleteService>,
    pub cloud_client: Option<Arc<dyn CloudDeleteClient>>,
    pub ontology_manager: Arc<OntologyManager>,
    pub search_orchestrator: Option<Arc<SearchOrchestrator>>,
    pub llm: Option<Arc<dyn Llm>>,
    pub transcriber: Option<Arc<dyn Transcriber>>,
    pub graph_db: Option<Arc<dyn GraphDBTrait>>,
    pub vector_db: Option<Arc<dyn VectorDB>>,
    pub thread_pool: Option<Arc<dyn CpuPool>>,
    pub embedding_engine: Option<Arc<dyn EmbeddingEngine>>,
    pub ontology_resolver: Option<Arc<dyn OntologyResolver>>,
    pub session_store: Option<Arc<dyn SessionStore>>,
    pub session_manager: Option<Arc<SessionManager>>,
    pub checkpoint_store: Option<Arc<dyn CheckpointStore>>,
    pub responses_client: Option<Arc<dyn ResponsesClient>>,
    pub notebook_runner: Option<Arc<dyn NotebookRunner>>,
}

Pre-initialized pipeline component handles shared across all P2 handlers.

Obtained from state.components().

Fields

database: Arc<DatabaseConnection>

SeaORM database connection (implements IngestDb, DeleteDb). The AclDb implementation is provided by the closed cognee-access-control crate and wired through Self::acl_db.

acl_db: Option<Arc<dyn AclDb>>

Optional ACL backend. None in pure-OSS builds (which do not enforce permissions). Wired by the closed cloud assembly to a newtype that implements AclDb over the shared DatabaseConnection.

storage: Arc<dyn StorageTrait>

File storage backend.

delete_service: Arc<DeleteService>

Fully configured DeleteService (with storage + DB wired).

cloud_client: Option<Arc<dyn CloudDeleteClient>>

Optional cloud delete proxy used by POST /api/v1/forget.

ontology_manager: Arc<OntologyManager>

Ontology manager (per-user file storage).

search_orchestrator: Option<Arc<SearchOrchestrator>>

Pre-built search orchestrator. None means HTTP search is unwired — handlers return SearchError {500, "Internal server error"}.

llm: Option<Arc<dyn Llm>>

Configured LLM adapter for /api/v1/llm/* handlers.

transcriber: Option<Arc<dyn Transcriber>>

Transcriber for audio document processing (Whisper). None when the configured LLM provider does not support audio transcription.

graph_db: Option<Arc<dyn GraphDBTrait>>

Knowledge-graph DB used by the visualize router.

vector_db: Option<Arc<dyn VectorDB>>

Vector DB handle required by cognee_core::TaskContext when the add / cognify / memify convenience functions route through pipeline::execute (LIB-06). None means the corresponding pipeline handlers surface a 500 / 409 envelope at runtime.

thread_pool: Option<Arc<dyn CpuPool>>

CPU pool used by cognee_core::TaskContext. Same routing notes as vector_db.

embedding_engine: Option<Arc<dyn EmbeddingEngine>>

Text embedding engine used by the cognify pipeline (chunks, entities, summaries). None means the cognify / update handlers surface a 500 envelope at runtime.

ontology_resolver: Option<Arc<dyn OntologyResolver>>

Ontology resolver passed into the cognify pipeline. None means the cognify / update handlers fall back to a pass-through NoOpOntologyResolver, matching the CLI default when no ontology_file_path is configured.

session_store: Option<Arc<dyn SessionStore>>

Backing store for session Q&A history — wires the session source of POST /api/v1/recall (Python _search_session, recall.py:146-208). None means session-source recall returns empty (matches Python’s is_available short-circuit at recall.py:170-171). Reuses the SessionStore re-exported by cognee-search to avoid pulling cognee-session into the crate’s non-dev dependency graph.

session_manager: Option<Arc<SessionManager>>

Manager for agent-trace sessions and the per-session graph context snapshot — wires the trace and graph_context sources of POST /api/v1/recall (Python _search_trace / _fetch_graph_context). None means both sources return empty.

checkpoint_store: Option<Arc<dyn CheckpointStore>>

Checkpoint store used by improve Stage 4 (sync_graph_to_session) to persist per-session high-water marks and avoid re-syncing old edges.

responses_client: Option<Arc<dyn ResponsesClient>>

OpenAI Responses API client — wires POST /api/v1/responses (Python get_responses_router.py). None means the handler returns 500 “responses client is not wired” until embedders populate it.

notebook_runner: Option<Arc<dyn NotebookRunner>>

Notebook cell execution backend used by POST /api/v1/notebooks/{notebook_id}/{cell_id}/run. None means the handler returns 501 — the same envelope it returned in Stage A before Stage B landed — preserving wire compatibility for embedders that don’t want to expose code execution.

Implementations

impl ComponentHandles

pub async fn formatted_graph_data( &self, dataset_id: Option<Uuid>, user_id: Uuid, ) -> Result<Value, Error>

Return the formatted knowledge-graph data for a dataset as the JSON shape {"nodes": [...], "edges": [...]}.

Wires to cognee_graph::get_formatted_graph_data when both a graph_db handle and a dataset_id are available. When either is missing — e.g. the server is running in test mode without backends — returns the empty-graph fallback {"nodes": [], "edges": []} so that the WS frame still has a valid shape.

Python parity: cognee.modules.graph.methods.get_formatted_graph_data.

Trait Implementations

impl Clone for ComponentHandles

fn clone(&self) -> ComponentHandles

Returns a duplicate of the value.

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

Performs copy-assignment from source.