neon-realtime
Skill108 starsChanged 17 days agoApache-2.0View on GitHub
Neon Realtime makes any SQL query live. Build realtime, reactive apps and agents on Lakebase Postgres with live SQL queries and reactivity primitives. Use when app UI or agent state must update live from database rows (dashboards, feeds, collaboration, multi-user or multi-device sync), for optimistic mutations, or to replace polling or other realtime sync systems. Triggers include "live query", "subscribe", "sync engine", "useLiveQuery", "TanStack DB", "optimistic updates", "@neon/realtime". For push that doesn't come from database rows (presence, typing, LLM streams), use neon-functions.
- Reads credentials
We split this skill into the rules it gives an agent
11 rules. About 0 are common, 4 are distinctive, and 7 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 10
FIRST: Use the parent neon skill for a Neon overview, getting started with Neon, Neon development best practices, and more.
- Very specific: line 12
If the neon skill is not installed, fetch it from https://neon.com/docs/ai/skills/neon/SKILL.md or install it with:
- Distinctive: line 164
Read references/queries.md before writing query or sealing code, including Kysely and server functions.
- Distinctive: line 167
Authorize in the query: Check the session before sealing, and join through your membership tables so revoked access drops out.
- Distinctive: line 168
A sealed query is a bearer token: Keep it out of URLs, logs and persistent storage, and never fill its parameters from unchecked input.
- Very specific: line 169
Check a query: before wiring it in: neon realtime subscribe "SELECT ..." prints the initial result, then each transaction's changes, until …
- Distinctive: line 174
Read references/reactivity.md before writing React or TanStack DB code.
- Very specific: line 186
Always pass awaitTxId() a timeout: it keeps waiting through reconnects and renewals, and never settles if the txid doesn't reach the query.
- Very specific: line 191
Read references/ssr.md before rendering live queries on the server, for Next.js and TanStack Start.
- Very specific: line 193
Create the client in a useState initializer rather than at module level, render initialData as live rows without a loading state, and never …
- Very specific: line 194
Keep row shapes identical on both sides.
≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
FIRST: Use the parent neon skill for a Neon overview, getting started with Neon, Neon development best practices, and more.
If the neon skill is not installed, fetch it from https://neon.com/docs/ai/skills/neon/SKILL.md or install it with:
neon skills -s neon -yNeon Realtime
Neon's Realtime makes any SQL query live. Subscribe in realtime to your Lakebase Postgres database. Develop apps and agents with end-to-end reactivity using type-safe live queries and optimistic mutations.
Realtime is new and in beta, so its APIs aren't in your training data. Start from the core pattern below, and read each section's reference file before writing that code.
How It Works
Realtime's layers compose: a live query engine inside Postgres, the Live SQL API (a WebSocket API), lower-level client SDKs (seal(), subscribe()), and reactivity primitives such as useLiveQuery.
- In trusted backend code, define a live query (any read-only SQL, including joins, aggregates, ordering and limits), seal it with
NEON_REALTIME_SECRETand expose it from an authenticated endpoint or server function. The client can't read or change a sealed query's SQL or parameters. - The client subscribes with the sealed query. It receives the full initial result, then each committed transaction's changes as one batch. Results stay consistent, but delivery of every change event isn't guaranteed.
- Writes go through your own API, which returns each write's
txidforawaitTxId().
Core Pattern
The guides' chat app, with messages authorized through channel_members. The backend seals the query and returns each write's txid:
import { createRealtime, rawSql } from "@neon/realtime/server"
import { drizzleAdapter } from "@neon/realtime-drizzle"
import { and, desc, eq, getTableColumns, sql } from "drizzle-orm"
import { Hono } from "hono"
import { HTTPException } from "hono/http-exception"
import { requireUser } from "./auth"
import { db } from "./db"
import { channelMembers, messages, type Message } from "./schema"
export const realtime = createRealtime({ secret: process.env.NEON_REALTIME_SECRET!, db: "neondb", adapter: drizzleAdapter() })
// Each selected key must match its column name
export function messagesQuery(userId: string, channelId: number) {
return db
.select(getTableColumns(messages)).from(messages)
.innerJoin(channelMembers, eq(channelMembers.channel_id, messages.channel_id))
.where(and(eq(channelMembers.user_id, userId), eq(messages.channel_id, channelId)))
.orderBy(desc(messages.created_at))
.limit(50)
}
// Or raw SQL
const rawMessagesQuery = (userId: string, channelId: number) => rawSql<Message>(
`SELECT messages.author_id, messages.body, messages.channel_id, messages.created_at, messages.id FROM messages
INNER JOIN channel_members ON channel_members.channel_id = messages.channel_id WHERE channel_members.user_id = $1 AND messages.channel_id = $2 ORDER BY messages.created_at DESC LIMIT 50`,
[userId, channelId]
)
// The id of the transaction that made the change, read in the write's own RETURNING
const txid = sql<string>`pg_current_xact_id()::text`
const app = new Hono()
.post("/api/v1/messages/live", async (c) => {
const user = await requireUser(c)
const { channelId } = await c.req.json<{ channelId: number }>() // validate input
return c.json(await realtime.seal({ query: messagesQuery(user.id, channelId) }))
})
.post("/api/v1/messages", async (c) => {
const user = await requireUser(c)
const { id, channelId, body } = await c.req.json<{ id: string; channelId: number; body: string }>() // validate input
// One statement. The membership row supplies channel_id and author_id, so only a member's message is inserted.
// List every column in table order, alias each sql field, and cast untyped params (they resolve to text)
const [sent] = await db.insert(messages).select(
db.select({
id: sql`${id}::uuid`.as("id"), channel_id: channelMembers.channel_id, author_id: channelMembers.user_id,
body: sql`${body}`.as("body"), created_at: sql`now()`.as("created_at"),
})
.from(channelMembers)
.where(and(eq(channelMembers.channel_id, channelId), eq(channelMembers.user_id, user.id)))
).returning({ txid })
if (!sent) throw new HTTPException(403, { message: "Not a member of this channel" })
return c.json({ kind: "committed", txid: sent.txid })
})The React frontend subscribes and sends optimistic messages:
import { createRealtimeClient, type SealedLiveQuery } from "@neon/realtime/client"
import { drizzleParsers } from "@neon/realtime-drizzle/client"
import { useLiveQuery } from "@neon/realtime-react"
import { startTransition, useCallback, useOptimistic } from "react"
import type { Message } from "../backend/schema"
import { fetchSealedMessages, sendMessage } from "./api"
// One client per app, because subscriptions share its connection. Wrap the app in <RealtimeProvider client={client}>
export const client = createRealtimeClient({ url: import.meta.env.VITE_NEON_REALTIME_URL, parsers: { ...drizzleParsers } })
const NO_MESSAGES: readonly Message[] = []
type Props = { sealedQuery: SealedLiveQuery<Message>; channelId: number; userId: string }
export function Messages({ sealedQuery, channelId, userId }: Props) {
const refreshQuery = useCallback(() => fetchSealedMessages(channelId), [channelId])
const { data, status, error, utils } = useLiveQuery(sealedQuery, { refreshQuery })
// Once the sent message syncs, show only the synced row
const [messages, addOptimistic] = useOptimistic(data ?? NO_MESSAGES, (current, message: Message) =>
current.some(({ id }) => id === message.id) ? current : [message, ...current]
)
function send(body: string) {
// The server sets author and timestamp but keeps this id
const message = { id: crypto.randomUUID(), channel_id: channelId, author_id: userId, body, created_at: new Date() }
startTransition(async () => {
addOptimistic(message)
try {
const result = await sendMessage(message) // maps channel_id to channelId
if (result.kind === "committed") await utils.awaitTxId(result.txid, 10_000)
} catch { /* show "Couldn't confirm the change" */ }
})
}
if (status === "error") return <p role="alert">Live updates stopped: {error.message}</p>
if (data === undefined) return <p>Loading messages</p>
return (
<form action={(form) => send(String(form.get("body")))}>
<input name="body" required />
<ul>{messages.map((message) => <li key={message.id}>{message.body}</li>)}</ul>
</form>
)
}Choose the Stack
A live query replaces fetching app data from an API or another sync system. Reuse existing query code, and use the highest layer that fits rather than wiring changes into app state by hand.
| Use | For |
drizzleAdapter() from @neon/realtime-drizzle, or kyselyAdapter() from @neon/realtime-kysely | Queries written in Drizzle or Kysely |
rawSql from @neon/realtime/server | Any other ORM, or none |
useLiveQuery from @neon/realtime-react | React. Binds one sealed query to a component's state |
realtimeCollectionOptions from @neon/realtime-tanstack | Many views over the same data, or an app on TanStack DB. Oversync one broader query into a collection, then derive local live queries from it |
client.subscribe() from @neon/realtime/client | Other frameworks |
realtime.subscribe() from @neon/realtime/server | Agents and workers, in trusted backend code |
Migrating from Electric: Realtime is a drop-in replacement for Electric's role in your stack. Live SQL queries replace shapes, and realtimeCollectionOptions() replaces electricCollectionOptions().
Setup
- New app:
neon bootstrap my-app --template realtime-react, thenneon deployinmy-app. - Enable it on the linked branch with
neon realtime enable --allowed-origin <app origin>, or declarepreview: { realtime: true }inneon.ts(neon config add realtime) and runneon deploy. Child branches inherit it. It's unavailable on Claimable projects until claimed, or outside the supported regions. - Tables need a stable replica identity; a primary key is enough.
- Environment:
neon env pull --service postgres --service realtimewritesDATABASE_URL,NEON_REALTIME_URLandNEON_REALTIME_SECRET. Expose only the URL to the browser, under your framework's public prefix (likeVITE_).urloptions takeNEON_REALTIME_URL;dbtakes the database name fromDATABASE_URL. - Install
@neon/realtimeand the packages chosen above. Use these SDKs, not the still-evolving WebSocket protocol.
Define and Seal Queries
Read references/queries.md before writing query or sealing code, including Kysely and server functions.
- Read-only and deterministic. A raw query is a single
SELECTwith no;and$1placeholders. It can't callnow()orgen_random_uuid(), so pass such values as parameters, wrapping every value except strings, numbers, booleans, bigints and null in apgParamhelper (pgParam.timestamptz(date)). - Authorize in the query. Check the session before sealing, and join through your membership tables so revoked access drops out. Any web framework auth works, including Managed Better Auth, but not RLS yet. See Authorization.
- A sealed query is a bearer token. Keep it out of URLs, logs and persistent storage, and never fill its parameters from unchecked input.
- Check a query before wiring it in:
neon realtime subscribe "SELECT ..."prints the initial result, then each transaction's changes, until you stop it. With--output json, each change prints a full snapshot. SetdebugModeoncreateRealtimefor full database errors (development only). - Sealed queries expire. Pass
refreshQueryto the bindings or callrenew(), sealing the same query and parameters, or the subscription errors. New parameters need a new sealed query and subscription. In Next.js, renew from a Route Handler, because server actions run one at a time.
Subscribe
Read references/reactivity.md before writing React or TanStack DB code. Read references/subscribing.md before subscribing directly with client.subscribe(sealed) (getSnapshot, onChange, onStateChange), or from agents and workers, which add url to createRealtime and await realtime.subscribe(query) to seal in-process.
- One database per client. The first sealed query binds the connection to its
db. - Select a stable, unique key for every row and key rows by it, including TanStack DB's
getKey: usually the primary key, the grouped columns for aggregates, or the combined keys for joins that repeat a row. Never userowId, which everyrowsupdate invalidates. erroris final, even afterrenew(). Fetch a fresh sealed query and subscribe again, with a new client if the connection failed. WatchonStateChangeon direct subscriptions;useLiveQuery()reports errors throughstatus, not by throwing or suspending.- When the session changes, leave with a full page load. It drops the Realtime client, any TanStack
DbClientand the previous user's rows. The hook retries a failed renewal every second, so stop renewing once one fails as signed out. Check on every renewal and write that the session's user is the one the page was loaded for, or a write from the old page lands in the new user's account. See When the session changes.
Optimistic Writes
Realtime has no write path, so writes keep your backend's usual auth. Read references/mutations.md before writing mutations, for each layer.
- Return the txid from the write itself:
.returning({ txid: sql<string>`pg_current_xact_id()::text` }), in one round trip on any Postgres driver. Write a batch in one statement, so there's one txid. Never opendb.transaction()just toselect pg_current_xact_id(): that's four round trips holding a pooled connection. When a write genuinely needs several statements, put the txid in theRETURNINGof a write that always changes a row. - Always pass
awaitTxId()a timeout: it keeps waiting through reconnects and renewals, and never settles if the txid doesn't reach the query. - A write that changes no rows has no txid, so skip
awaitTxId()and let the optimistic state drop.
Server-Side Rendering
Read references/ssr.md before rendering live queries on the server, for Next.js and TanStack Start.
- Create the client in a
useStateinitializer rather than at module level, renderinitialDataas live rows without a loading state, and never share a TanStackDbClientbetween requests. - Keep row shapes identical on both sides.
JSON.stringifyturns dates into strings, andint8andnumericarrive as strings.
More Docs
| Task | Page |
| Beta scope, regions, pricing, FAQ, other systems | Overview |
| Architecture and consistency | How it works |
| A first live query | Quickstart |
| Every SDK export | TypeScript SDK reference |
| Full-stack apps with auth and SSR | Next.js, TanStack |
| Live agent workflows | Mastra |
| ORM adapters | Drizzle, Kysely |
In this repository9 other files it gives its agents.
- AGENTS.mdAGENTS.md
- neon-ai-gatewaySkill
- neon-authSkill
- neon-functionsSkill
- neon-object-storageSkill
- neon-postgres-branchesSkill
- neon-postgres-egress-optimizerSkill
- neon-postgresSkill
- neonSkill
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

