doc for persistence? like what ai sdk have?

[ManuRodgers]

https://github.com/vercel-labs/ai-sdk-persistence-db/tree/main

[smorimoto]

@AlemTuzlak Do you have any thoughts on this? The persistence features around LangGraph are incredibly well-designed out and really impressive. That said, I don’t particularly resonate with their approach. It makes me think how great it would be if TanStack AI supported persistence and checkpointing features like this!
https://docs.langchain.com/oss/javascript/langgraph/persistence

[willwillems]

Honestly, some sort of recommendation around persistence would be extremely helpful. I'm currently passing along a conversation ID as context and using tee to split the streamed response and use one of those to aggregate again and store. I don't mind the stateless approach this lib takes but almost anyone will need persistence and currently there's just a finish handler on the front end as an obvious hook?

[DiegoGBrisa]

I just opened a concrete API proposal for this: #374

The idea is a simple ChatPersistenceAdapter interface on ChatClient with getItem/setItem/removeItem — same shape as the existing ToolCacheStorage. Framework wrappers just pass it through. Would love your thoughts on the API shape there.

[willwillems]

This is what I've been using:

/**
 * Yields every item from source transparently while collecting them.
 * After the source exhausts, calls `onComplete` with all collected items
 * within the iteration lifecycle, before signaling done.
 *
 * This ensures the consumer (e.g., an HTTP response stream) stays open
 * until `onComplete` finishes, binding post-stream work to the
 * response lifetime rather than leaving it as a detached promise.
 */
export async function* withOnComplete<T>(
  source: AsyncIterable<T>,
  onComplete: (items: T[]) => Promise<void>,
): AsyncGenerator<T> {
  const items: T[] = []
  for await (const item of source) {
    items.push(item)
    yield item
  }
  await onComplete(items)
}

and just wrap chat() in that.

[AlemTuzlak]

You guys deserve so much more than just docs for persistence! Our next big release will be headless perisstence layer built on top of all the current API's so you can bring your own persistence with a few lines of code!