Skip to main content

Optimistic Updates

Even though Convex queries are completely reactive, sometimes you'll want to update your UI before the mutation changes propagate back to the client. To accomplish this, you can configure an optimistic update to execute as part of your mutation.

Optimistic updates are temporary, local changes to your query results which are used to make your app more responsive. These updates are made by functions registered on a mutation invocation with the .withOptimisticUpdate configuration option.

Optimistic updates are run when a mutation is initiated, rerun if the local query results change, and rolled back when a mutation completes.

Optimistic updates are in beta

Optimistic updates are currently a beta feature. If you have feedback or feature requests, let us know on Discord!

Simple example

Here is how an optimistic update could be added to an increment mutation in a simple counter app:

import { api } from "../convex/_generated/api";
import { useMutation } from "convex/react";

export function IncrementCounter() {
const increment = useMutation(api.counter.increment).withOptimisticUpdate(
(localStore, args) => {
const { increment } = args;
const currentValue = localStore.getQuery(api.counter.get);
if (currentValue !== undefined) {
localStore.setQuery(api.counter.get, {}, currentValue + increment);

const incrementCounter = () => {
increment({ increment: 1 });

return <button onClick={incrementCounter}>+1</button>;

Optimistic updates receive a localStore, a view of the Convex client's internal state, followed by the arguments to the mutation.

This optimistic update updates the api.counter.get query to be increment higher if it's loaded.

Complex example

If we want to add an optimistic update to a multi-channel chat app, that might look like:

import { api } from "../convex/_generated/api";
import { useMutation } from "convex/react";
import { Id } from "../convex/_generated/dataModel";

export function MessageSender(props: { channel: Id<"channels"> }) {
const sendMessage = useMutation(api.messages.send).withOptimisticUpdate(
(localStore, args) => {
const { channel, body } = args;
const existingMessages = localStore.getQuery(api.messages.list, {
// If we've loaded the api.messages.list query, push an optimistic message
// onto the list.
if (existingMessages !== undefined) {
const now =;
const newMessage = {
_id: crypto.randomUUID() as Id<"messages">,
_creationTime: now,
localStore.setQuery(api.messages.list, { channel }, [

async function handleSendMessage(
channelId: Id<"channels">,
newMessageText: string,
) {
await sendMessage({ channel: channelId, body: newMessageText });

return (
<button onClick={() => handleSendMessage(, "Hello world!")}>
Send message

This optimistic update changes the api.messages.list query for the current channel to include a new message. The newly created message object should match the structure of the real messages generated by the api.messages.list query on the server.

Because this message includes the client's current time (not the server's), it will inevitably not match the api.messages.list query after the mutation runs. That's okay! The Convex client will handle rolling back this update after the mutation completes and the queries are updated. If there are small mistakes in optimistic updates, the UI will always eventually render the correct values.

Similarly, the update creates a temporary Id with new Id("messages", crypto.randomUUID()). This will also be rolled back and replaced with the true ID once the server assigns it.

Lastly, note that this update creates a new array of messages instead of using existingMessages.push(newMessage). This is important! Mutating objects inside of optimistic updates will corrupt the client's internal state and lead to surprising results. Always create new objects inside of optimistic updates.

Learning more

To learn more, check out our API documentation:

If you'd like some hands on experience, try adding optimistic updates to the tutorial app! If you do, you should notice the app feels snappier — just a little, Convex is pretty fast already! — but otherwise works the same.

To explore even further, try inserting a mistake into this update! You should see a flicker as the optimistic update is applied and then rolled back.