Chat persistence for the Vercel AI SDK

The route

Same behaviour, one call

Both of these are in the docs and both typecheck against the published package on every build. The left one is what the persistence pattern asks you to maintain per app.

const { id, messages } = await req.json();

const existing = await store.loadMessages(id);
const known = new Set(existing.map((m) => m.id));
const fresh = messages.filter((m) => m.role === "user" && !known.has(m.id));
if (fresh.length > 0) await store.appendMessages(id, fresh);

const history = [...existing, ...fresh];
const result = streamText({
  model: openai("gpt-5"),
  messages: await convertToModelMessages(history),
});

let persisted = false;
const persist = async ({ responseMessage }) => {
  if (persisted || responseMessage.parts.length === 0) return;
  persisted = true;
  await store.appendMessages(id, [responseMessage]);
};

return result.toUIMessageStreamResponse({
  generateMessageId: generateId,
  onEnd: persist,
  onFinish: persist,
});

export const POST = chatHandler({
  store,
  execute: ({ modelMessages }) =>
    streamText({
      model: openai("gpt-5"),
      messages: modelMessages,
    }),
});

Branching

Regenerate, and the old answer is still there

Every message points at its parent, so a regenerated reply is a sibling rather than an overwrite - and each sibling keeps the turns that followed it. Switch the active leaf and the whole conversation below it comes back. Nothing is ever deleted.

siblingsOf

Every variant of one message, oldest first, with the index of the live one.

forkAt

Edit a question and the old version and its replies stay, on their own branch.

sdk_version

Stamped per row, so an AI SDK major bump is a migration rather than a guess.

An illustration of the stored shape - the buttons are siblingsOf plus setActiveLeaf, and the AI SDK itself still has no answer for the tree (vercel/ai#2929, open since 2024). Run it against a real database in your browser.

Coverage

What ai-sdk-threads gives you

One-line chat route

chatHandler replaces the load, store, stream, store boilerplate every AI SDK app writes by hand.

Branching

Edit or regenerate and the old version survives as a sibling, the way ChatGPT does it.

Resumable streams

resumableChat ships the POST, GET and DELETE trio, so a reload mid-answer picks the stream back up.

Postgres or SQLite

One ThreadStore contract over either, verified by a parity suite run against both.

UIMessage-native

Parts and metadata stored verbatim, never flattened to a content string that loses tool calls.

Zero runtime dependencies

ai, drizzle-orm and resumable-stream are peers, the last two optional. Install only what you use.

Keyset pagination

listThreads pages by cursor rather than OFFSET: one query per page, and a page 50,000 rows deep measured 1.13x the first.

Edge-safe core

No Node globals anywhere in src, enforced by a second typecheck that compiles without Node types.

Requirements

What your project needs

Alternatives

Compared to the alternatives

marks the tool ahead on that row, whichever tool it is

Nothing here is a like-for-like competitor, which is rather the point. If you want the managed experience, assistant-ui and Convex are good at it. Everything this package does today is MIT and stays that way; only a future managed layer could ever be otherwise. This exists for the case where the conversation has to stay in a database you control, and where regenerate and edit need to survive a reload. If you started from the Vercel template, its tables import straight across.

The project

By the numbers

245

downloads a month

1.04 kB

minified + gzipped

113 kB

unpacked size

1

GitHub star