Tracking 24 creators · live now

Streams are chaos.
We gave them clarity.

Your real-time Twitch operations workspace. Track Kai Cenat and your full watchlist with live telemetry, ML-powered peak forecasts, and an embedded player — all without leaving the room.

Live viewer telemetryMulti-streamer watchlistPeak forecastingZ-score anomaly detectionEmbedded Twitch playerDiscord go-live hooksGroup leaderboardsSignal Lab heatmapsPerformance timelinesSQLite session historySearch & add channelsLive viewer telemetryMulti-streamer watchlistPeak forecastingZ-score anomaly detectionEmbedded Twitch playerDiscord go-live hooksGroup leaderboardsSignal Lab heatmapsPerformance timelinesSQLite session historySearch & add channels

01 / Capabilities

A control room
for going live.

Built for operators who need to see, track, and forecast streamer activity at a glance. Every panel is composed for one job: turn raw Twitch firehose into a readable workspace.

Live telemetry, refreshed every 60 seconds.

Cached dashboard reads so a single refresh powers every panel without re-spamming Twitch. Status, viewer counts, uptime, and category data feed straight into the workspace.

Now+30 min projection
→ /api/dashboard

Search & add any channel.

Hero-level search hits the Twitch API and adds new creators directly to your tracked watchlist — no JSON config, no restart.

→ /api/search
σ

Anomaly watch.

Z-score detection over viewer-count diffs flags sudden surges. Threshold defaults to 2.5σ — configurable per deployment.

σ > 2.5 → alert

ML peak forecasts — with diagnostics, not just a number.

A degree-2 polynomial regression baseline projects the next 30 minutes. Output includes confidence label, MAE, naive-baseline MAE, and a standard-error band so you can trust (or ignore) the call.

→ /api/ml/predict/<login>

Embedded player.

Watch the focused channel inside the workspace using Twitch’s official embed — no second tab, no context loss.

live · in-app

Discord alerts.

Pipe go-live transitions, milestone crossings, and category swaps into a webhook channel of your choice.

webhook hook

Group leaderboards.

Live viewers, historical peaks, session activity, and momentum — ranked across the whole watchlist or a focused cohort.

→ /api/analytics/leaderboards

Signal Lab analytics.

Bring TwitchTracker-style context into the workspace: market pulse cards, top streams, creator performance, timeline growth, category mix, and activity heatmaps.

→ /api/analytics/signal-lab/<login>

Local SQLite history.

Every snapshot, session, and chat sentiment writes to a thread-safe local DB. Drill into one creator's last 100 datapoints in milliseconds.

stream_snapshots · events
60sDefault poll cadence
15Public API routes
±1.96σForecast confidence band
0/kbTwitch credentials in repo

02 / Workflow

From watchlist
to verdict.

Three short steps separate dropping in a creator from getting a forecast and an alert. Everything runs locally against your own Twitch developer credentials.

STEP 01

Track

Drop in a Twitch login or search by display name. New creators land on the watchlist with full history scaffolding from the first poll.

POST /api/watchlist
GET /api/search?q=<name>
STEP 02

Read

Live status, viewer trace, category, uptime, and historic peaks update on the cadence. The cache layer makes one refresh feel like ten.

GET /api/dashboard
GET /api/stream/<login>
STEP 03

Forecast

Project the next 30 minutes with confidence diagnostics and a baseline MAE comparison. Anomaly flags fire when σ crosses 2.5.

GET /api/ml/predict/<login>
GET /api/analytics/overview

03 / The Workspace

One creator, centered.
Whole watchlist, in periphery.

The Audience Operations Workspace is a three-column layout: a watchlist rail on the left, the focused creator and embedded player in the middle, and the live event wire on the right. Designed for one read, every refresh.

kc-live.local / dashboard
K

Kai Cenat

@kaicenat · 4h 12m uptime
● LIVEJust Chatting
Audience trace · last 30 min
247,832+18% momentum
Predicted peak
312,400
Confidence
High
Momentum
+18%
Best peak
428k

04 / Forecast

A forecast that tells you
when to trust it
.

Most "AI" chart projections give you a confident-looking line and nothing else. Ours ships with the diagnostics that decide whether the line is worth reading.

Peak prediction

Polynomial baseline,
honest about itself.

A degree-2 polynomial regression projects the next thirty minutes from the last session's trace. Confidence is a function of the standard error band, current viewer scale, and whether we’re actually beating a naive last-value baseline.

  • 01Confidence label. High, medium, or low — based on relative SE and baseline win.
  • 02Model MAE vs naive MAE. If the baseline beats us, we say so.
  • 03±1.96σ band. Upper and lower bounds for every horizon point.
  • 04Trend classifier. Growing, stable, or declining versus current.
/api/ml/predict/kaicenatConfidence: high
Model MAE1,840
Baseline MAE3,210
TrendGrowing

Anomaly watch

Spikes named
before they matter.

A z-score sweep over viewer-count diffs flags every datapoint that breaks past 2.5σ from recent volatility. Surges, raids, and milestone moments surface in the event wire and trigger your Discord webhook automatically.

  • σConfigurable threshold via ALERT_THRESHOLDS
  • Stored on every snapshot — queryable via the analytics overview endpoint
  • Surfaces in event wire and Discord webhook
Open the workspace
kaicenat · last 60 minutes
z-score 3.4
Z 3.4Surge spike at 13:42 UTC — +18.4% on viewer diff+44k
Z 2.7Category swap → Just Chatting at 13:11 UTC+12k
Z 2.6Raid-shaped pulse at 12:48 UTC+9.8k

05 / Signal Lab

TwitchTracker depth,
KC Live speed.

The new Signal Lab tab adds market summaries, top-stream tables, creator performance cards, over-time trend charts, category mix, and calendar heatmaps without leaving the operations desk.

MARKET

Global pulse

Viewers watching, live channels, unique games, archive peaks, and top live channels roll up from the watchlist and local history.

GET /api/analytics/signal-lab/<login>
CREATOR

Performance summary

Hours streamed, average viewers, peak viewers, hours watched, audience lift, games streamed, and active days are grouped by time range.

RANGES 7d · 30d · 3m · all
HEATMAP

Activity memory

A calendar grid shows which days actually carried the channel, with intensity based on streamed minutes and peak activity.

SOURCE SQLite snapshots · sessions

06 / API

Every panel is
a real route.

Fifteen public endpoints power the workspace. Hit them yourself for custom integrations, bots, dashboards, or notebook analyses — same data, same cache.

GET   /api/health
GET   /api/config
GET   /api/dashboard
GET   /api/stream/<login>
GET   /api/history/<login>
GET   /api/ml/predict/<login>
GET   /api/search?q=<query>
POST  /api/watchlist
DEL   /api/watchlist/<login>
GET   /api/analytics/overview
GET   /api/analytics/leaderboards
GET   /api/analytics/compare?logins=...
GET   /api/analytics/stream/<login>
GET   /api/analytics/signal-lab/<login>
GET   /

Ready to see the room?

Open the workspace. Drop in your watchlist. Watch one read replace ten browser tabs.

© KC Live · 2026 · Audience Operations Workspace