Twitch Live Stream Intelligence

Monitor Twitch streams, detect viewer growth, track live channels, discover rising categories, and collect structured Twitch intelligence automatically — from public data only.

This is not a generic scraper. It turns publicly available Twitch information into structured intelligence: it tells you who is live now, how their audience is moving, which categories are heating up, and what brands are being mentioned — and it detects meaningful changes over time when you run it on a schedule.

Responsible use: This Actor only reads publicly available Twitch data. It does not create viewers, inflate counts, automate accounts, generate fake engagement, or bypass any authentication, CAPTCHA, or anti-bot protection. It is a monitoring and analytics tool.

---

Who it's for

• Creators & managers — track your own and competitors' live performance and growth.
• Agencies & talent scouts — discover rising streamers and trending categories.
• Brands & marketers — surface brand/sponsorship mentions across live streams.
• Researchers & analysts — build your own historical dataset of the live Twitch ecosystem.

What it does

• Detects whether each channel is live or offline.
• Collects stream title, streamer, category/game, viewer count, follower count, start time, duration, language, tags, thumbnail, profile image, and URLs.
• Discovers currently-live channels by category/game.
• Computes viewer change, growth %, velocity, peak/min/average viewers, and flags trends as normal, rising, spike, or declining.
• Emits structured events: stream_start, stream_end, title_change, category_change, language_change, viewer_change.
• Aggregates category intelligence (live channel count, total/average/median viewers, top channels).
• Produces streamer intelligence (latest snapshot + trend per channel).
• Detects brand & sponsorship signals from public titles/tags/description — labelled as detected mentions, never confirmed sponsorships.

Input

Run it with just a channel name. All fields:

Field	Type	Default	Description
channels	array	–	Usernames or URLs, e.g. "xqc" or "https://twitch.tv/xqc".
categories	array	–	Games/categories to discover live channels in, e.g. "Just Chatting".
languages	array	–	Keep only these languages (2-letter, e.g. "EN").
liveOnly	boolean	false	Skip offline channels.
minimumViewers / maximumViewers	integer	0	Viewer bounds for live streams (0 = no bound).
minimumStreamDurationSeconds	integer	0	Minimum uptime for live streams.
monitoringIntervalSeconds	integer	0	Re-check every N seconds within one run (0 = single snapshot).
maximumMonitoringDurationSeconds	integer	0	Cap for a monitoring run.
maximumChannels	integer	0	Safety cap per cycle (0 = no cap).
categoryStreamsPerGame	integer	30	How many live streams to pull per category.
enableAnalysis	boolean	true	Compute growth/category/streamer/brand intelligence.
previousDatasetId	string	–	Dataset ID of a prior run — used as baseline for cross-run change detection.
proxyConfiguration	object	Apify Proxy	Proxy for Twitch requests.

Example inputs

Single streamer

{ "channels": ["xqc"] }

Monitor ten competitors

{ "channels": ["pokimane","kaicenat","ninja","shroud","amouranth","tarik","summit1g","hasanabi","xqc","ludwig"], "liveOnly": true }

Track a gaming category and find rising channels

{ "categories": ["League of Legends"], "minimumViewers": 500, "languages": ["EN"], "enableAnalysis": true }

Continuous monitoring (build a short series in one run)

{ "channels": ["kaicenat"], "monitoringIntervalSeconds": 60, "maximumMonitoringDurationSeconds": 600 }

Output

1) Channel observations (default dataset)

One normalized record per channel per cycle:

{
  "channelName": "eliasn97",
  "streamerName": "eliasn97",
  "channelUrl": "https://www.twitch.tv/eliasn97",
  "streamUrl": "https://www.twitch.tv/eliasn97",
  "isLive": true,
  "title": "Reactions & Talks | !iconleague",
  "category": "Just Chatting",
  "gameId": "509658",
  "viewerCount": 25454,
  "followerCount": 2412771,
  "language": "DE",
  "tags": ["Deutsch"],
  "startedAt": "2026-08-30T17:01:30Z",
  "observedAt": "2026-08-30T17:35:02.000Z",
  "streamDurationSeconds": 2012,
  "thumbnailUrl": "https://static-cdn.jtvnw.net/previews-ttv/live_user_eliasn97-1920x1080.jpg",
  "profileImageUrl": "https://static-cdn.jtvnw.net/.../profile_image-300x300.png",
  "viewerChange": 320,
  "viewerChangePercent": 1.28,
  "growthClassification": "normal",
  "detectedBrands": [{ "brand": "Red Bull", "confidence": 0.6, "sources": ["title"], "detectedMention": true, "confirmedSponsorship": false }],
  "sponsorshipSignals": [{ "source": "title", "phrase": "sponsored by" }],
  "detectedPromoCodes": [{ "source": "title", "code": "ELIAS10" }],
  "dataSource": "twitch-public-gql"
}

2) Stream events (named dataset events)

{ "channelName": "kaicenat", "eventType": "viewer_change", "observedAt": "2026-08-30T17:35:02.000Z", "previousValue": 40000, "newValue": 62000, "metadata": { "viewerChange": 22000, "viewerChangePercent": 55, "classification": "spike" } }

3) Category intelligence (named dataset category-intelligence)

{ "category": "Just Chatting", "liveChannelCount": 12, "totalObservedViewers": 184300, "averageObservedViewers": 15358, "medianObservedViewers": 9200, "topChannels": [{ "channelName": "eliasn97", "viewerCount": 25454, "title": "..." }], "observedAt": "2026-08-30T17:35:05.000Z" }

Additionally, a run summary is written to the key-value store record OUTPUT, and streamer intelligence (latest + trend per channel) to STREAMER_INTELLIGENCE.

Scheduling & historical data

Schedule the Actor (e.g. every 5–15 minutes). Each run appends observations you can combine into a historical dataset. To detect changes across runs, pass the previous run's dataset ID as previousDatasetId — the Actor uses each channel's most recent prior observation as the baseline for viewer changes and events.

Reliability & error handling

• Batched GraphQL requests with retries + exponential backoff and rate-limit awareness.
• One unavailable/banned/renamed channel is skipped and logged — it never fails the run.
• Deterministic, consistently-typed output for easy downstream joins.

Limitations

• Viewer/follower counts reflect publicly visible values at observation time — they are point-in-time samples, not Twitch's official historical analytics.
• Brand/sponsorship results are detected mentions from public text, not confirmed commercial relationships.
• Category discovery returns a sample of top live streams per category (categoryStreamsPerGame), not the entire category.

Architecture

Modular by design — twitch (collection), normalize, compare (history/events), analytics, state (cross-run), and main (orchestration/output) — so future capabilities (Slack/Discord/email alerts, dashboards, creator ranking, historical DB, e-commerce product matching) drop in cleanly.