ai-core/client-persistence

Pass a persistence adapter like localStoragePersistence() to the useChat persistence option along with a stable threadId. The adapter stores the full transcript per threadId, so a reload repaints the conversation and restores any pending interrupt.

localStoragePersistence is synchronous and survives restarts but is quota-bound; indexedDBPersistence handles larger data asynchronously; sessionStoragePersistence only survives same-tab reloads. For large transcripts, multi-device sync, or compliance, use persistence: true with a server store instead.

Client storage adapters like localStorage are per-browser, so they cannot sync across devices. Use server-authoritative mode (persistence: true) with withPersistence and a hydrate route on the server so any device can reload the thread by threadId.

persistence: true caches nothing in the browser — it hydrates from the server by threadId on mount. If the server has no history for that threadId (no withPersistence or reconstructChat route), the chat loads empty.

Only if the route uses delivery durability via toServerSentEventsResponse(stream, { durability: ... }). Persistence alone restores finished runs and pending interrupts; rejoining an in-flight stream requires the durable stream so the client can call joinRun.

No. Generation hooks accept only persistence: true (boolean), which hydrates the last generation for a required threadId from the server on mount. Nothing is cached in the browser, and media bytes restore only if the server persists artifacts with an artifactUrl mapper.