Struct SubagentJsonl 

pub struct SubagentJsonl { /* private fields */ }

Adapter for subagent JSONL transcripts.

Each line in the transcript is a TranscriptEntry. The adapter produces one IngestDocument per entry whose flat text content is non-empty. Context for each entry is the plain text of the context_window preceding entries.

Examples

use zeph_memory::graph::ingest::{SubagentJsonl, IngestSourceAdapter, ImportBatchId};

let raw = r#"{"seq":0,"timestamp":"2026-01-01T00:00:00Z","message":{"role":"user","content":"hello","parts":[]}}"#;
let adapter = SubagentJsonl::new("task-42");
let batch = ImportBatchId::new();
let docs = adapter.parse(raw, &batch).unwrap();
assert!(!docs.is_empty());

Implementations

impl SubagentJsonl

pub fn new(task_id: impl Into<String>) -> Self

Creates a new SubagentJsonl adapter for the given task ID.

task_id is embedded in every document’s source_uri as "subagent:<task_id>#<seq>".

The context_window defaults to 3 preceding messages.

pub fn with_context_window(self, n: usize) -> Self

Overrides the context window size.

Trait Implementations

impl IngestSourceAdapter for SubagentJsonl

fn parse( &self, raw: &str, batch_id: &ImportBatchId, ) -> Result<Vec<IngestDocument>, MemoryError>

Parse a JSONL transcript string.

Lines that fail to parse are skipped with a warning; entries whose flat text content is empty are also skipped (nothing to extract).