import { createClient } from "@supabase/supabase-js";
const supabase = createClient("https://api.busymate.net", PUBLISHABLE_KEY, {
global: { headers: { Authorization: `Bearer ${oauthToken}` } },
});
supabase
.channel("ws:<workspace_id>")
.on("broadcast", { event: "entry" }, ({ payload }) => {
console.log("new request", payload.method, payload.url, payload.status);
})
.subscribe();Push,
端到端推送驱动。流量、状态或审计一有变化，就在你自己的代码里订阅到。
记录消防水龙带
每个抓到的请求落地瞬间即被广播——就在仪表盘自己消费的那条频道上。你的监控与团队同一瞬间看到同样的东西。
表变更即流
设备、设置、标签、服务组、共享待办板——全部以 postgres_changes 事件送达。只要数据库里变了，你就会知道。没有定时器，没有刷新循环。
不止于流量
实时审计日志、记忆导入进度、设备控制事件也都可订阅。构建告警机器人、大屏看板，或响应平台本身的自动化。
从文档里直接订阅
仪表盘内的 WS 浏览器用你自己的令牌加入真实频道，并在事件到达时美观打印。写下第一行代码之前，先看看这条流。
Subscribe once,
Realtime-not-polling is a standing engineering rule here — every feature is push-driven end to end, and the dashboard itself never polls.
- 01
Connect once
One WebSocket to api.busymate.net's Realtime service, authenticated with the same OAuth token as REST and MCP.
- 02
Join channels
Subscribe to topics — a workspace firehose, a single device, fleet-wide status — each authorized by the same row-level security as the tables themselves.
- 03
Events push to you
Database triggers broadcast changes the instant rows land. No polling loop, no missed window between polls — the push IS the source of truth arriving.
- 04
React — or publish back
Feed alerting, wall dashboards and bots — or drive control flows the other way, like continuing a paused breakpoint over the proxy-control channel.
// Raw table changes — the shared to-do board
supabase
.channel("todos:all")
.on(
"postgres_changes",
{ event: "*", schema: "public", table: "todos" },
(change) => console.log(change.eventType, change.new),
)
.subscribe();A map of the
The channels your code can join — the same canonical list the in-dashboard WS explorer documents and subscribes.
ws:<workspace_id>broadcastThe workspace entries firehose — every captured request from every device, the moment it lands. The exact channel the dashboard feed renders.
ws:<workspace_id>:<owner_user_id>broadcastThe same firehose narrowed to one owner's devices — what a non-operator account subscribes.
device:<uuid>broadcastPer-device control plane: settings pushes, VPN and unpair commands, browser and farm command round-trips, live-activity messages.
devices:alldbDevice renames, pairing and status changes, fleet-wide — the reason device lists everywhere update without a refresh.
proxy-controlbroadcastDashboard ↔ proxy control flow: continue a request paused at a breakpoint, or resend a captured request.
settings:global · settings_device:all · settings_user:alldbSettings pushes at every tier — capture engines refetch their effective settings the moment any layer changes.
service_groups:alldbService-group membership and rule changes, so grouped devices re-apply rules live.
breakpoint_events:alldbEvery request pausing and resuming at a breakpoint, as it happens — the read side of the continue flow.
audit:all · audit:<actor_uuid>dbThe live audit trail — every action by every human and agent, fleet-wide for operators or scoped to your own events.
todos:allpostgres_changesThe shared to-do board as raw table-change events — the simplest channel to start with.
React in
Alerting bots
Watch the firehose for error spikes on a production host and page your team the second they start — not on the next poll.
Wall dashboards
Render live traffic, device fleet state and the audit trail on a NOC screen — the same push the product dashboard consumes.
Reactive automation
React to a device going offline, a breakpoint pausing, or a settings change — then act back through MCP or the proxy-control channel.
Push is one of
Every capability aligns across the dashboard, MCP, REST, WebSockets and BusyBro — a standing rule, checked on every ship. Read over REST, subscribe over WS, act over MCP: one identity, one permission model.
Before you
What do I connect with?
supabase-js is the shortest path (it handles the Realtime protocol, auth and reconnection), but any client that speaks the Supabase Realtime protocol over WebSockets works. Authenticate with the same OAuth token as REST and MCP.
Broadcast vs postgres_changes — what's the difference?
Most channels are database-triggered broadcasts: a trigger fans a shaped payload to a named topic the instant a row lands — scalable and authorized per topic. A few simple tables (like the to-do board) stream raw postgres_changes events instead.
Can I publish, or only subscribe?
Both. Subscriptions cover the feeds; control flows publish back — the proxy-control channel carries breakpoint-continue and resend-request, and the confirm-gated MCP tools publish device commands on your behalf.
Will I see other people's traffic?
No — channel access is authorized by row-level security on the Realtime layer. Plain accounts get their own devices' feed; the fleet-wide firehose and audit stream require operator capabilities.
Why not just poll the REST API?
Realtime-not-polling is a standing engineering rule here: every feature is push-driven end to end, and the dashboard itself never polls. Push gives you lower latency, no missed events between polls, and no wasted queries.