The open-source agent harness - the runtime layer that turns an LLM into a working agent.
import type { AgentChatServer } from './server/types.js';
import { buildSnapshotFromSessionEvents } from './convertTurnMessages.js';
import { getSession } from './sessions.js';
import type { SessionSnapshot } from './sessionSnapshot.js';
const inflightBySessionId = new Map<string, Promise<SessionSnapshot>>();
/**
* Loads a session snapshot once per concurrent burst for a given session id.
* React Strict Mode and overlapping fetch/load paths share the same in-flight
* request instead of duplicating getSession + listEvents calls.
*
* `onProgress` is called after each complete turn is ingested so the caller
* can update the UI progressively while history is being processed.
*/
export function loadSessionSnapshot(
server: AgentChatServer,
sessionId: string,
onProgress?: (snap: SessionSnapshot) => void,
): Promise<SessionSnapshot> {
let inflight = inflightBySessionId.get(sessionId);
if (inflight == null) {
inflight = getSession(server, sessionId)
.then(() => buildSnapshotFromSessionEvents(server, sessionId, onProgress))
.finally(() => {
if (inflightBySessionId.get(sessionId) === inflight) {
inflightBySessionId.delete(sessionId);
}
});
inflightBySessionId.set(sessionId, inflight);
}
return inflight;
}