yt-view-bot

本页保存的是公开项目资料快照，阅读过程不需要连接 GitHub。

---

🧭 Table of Contents

• Overview
• What is a YouTube Viewer Bot?
• Key Features
• The Problem
• The Solution
• Quick Start
• Module Catalog
• Session & Proxy Management
• Retention & Watch-Time Engine
• System Requirements
• Installation
• AI Integration (2026)
• How It Works
• Comparison
• Known Issues
• Changelog
• FAQ
• Community & Support

---

📊 Overview

Category	Details
Product type	Windows desktop application (compiled .exe)
Purpose	Automated YouTube view generation with retention simulation
Primary interface	GUI dashboard + headless batch mode
Concurrency model	Session-pool based, proxy-rotated
Watch-time engine	Per-session timeline scripting with drift correction
Proxy support	HTTP, SOCKS4, SOCKS5, residential, mobile
Data persistence	Local SQLite job store, encrypted credential vault
Distribution	Direct download from project landing page

The yt-view-bot application is a session-orchestration framework built specifically for YouTube view and watch-time automation. It manages pools of browser sessions, rotates proxy identities, and scripts realistic playback timelines (including seek, pause, and skip events) to reproduce the watch patterns a genuine viewer produces. Everything runs locally from a single .exe — no runtime installs, no Python environment, no dependency chain.

---

🤖 What is a YouTube Viewer bot?

Term	Explanation
Session	An isolated browser profile (cookies, UA, viewport, fingerprint) representing one synthetic viewer
Proxy pool	Rotating set of IP endpoints assigned per-session so views don’t share an origin
Retention script	A timeline of play/pause/seek actions that mimics human watch behavior
Drift	The gap between wall-clock playback time and scripted timeline position; corrected periodically
View-count boundary	The threshold at which YouTube may freeze or filter counts; the bot throttles to stay under it
Fingerprint	Composite of canvas hash, WebGL renderer, fonts, timezone, and UA sent to YouTube
Job	One configured run — N sessions, M views, one target video, one proxy pool

The bot is not a “click play N times” loop. It manages the full identity of each session — fingerprint, cookie jar, and proxy binding are locked together at spawn time and cannot drift mid-run. This matters because YouTube’s view validation checks consistency between IP, cookie state, and playback telemetry.

---

🎯 Key Features

Feature	Description	Benefit
Fingerprint locking	Canvas/WebGL/UA/fonts sampled once and reused per session	Views don’t get grouped as one synthetic actor
Proxy rotation modes	Round-robin, sticky, per-session, per-job	Fine control over IP exposure for a target video
Retention scripting	Timeline JSON with play, pause, seek, skip, mute	Watch-time looks organic instead of a 100% full play
Drift correction	Re-syncs timeline against actual currentTime every 4s	Prevents runaway desync on slow proxies
Headless batch mode	CLI flag runs jobs without GUI	Schedulable via Windows Task Scheduler
Per-job concurrency cap	Limits simultaneous sessions (default 3)	Reduces proxy burn and detection surface
Encrypted vault	AES-encrypted proxy creds and cookie jars at rest	No plaintext credentials on disk
Live telemetry panel	Real-time view-count delta, session health, proxy latency	You see failures before the job finishes
Resume-after-crash	Job state persisted to SQLite; resume from last completed session	Long jobs survive reboots
Warm-up mode	Sessions browse homepage/related videos for 30–90s before target	Reduces “instant-jump” signal
AI retention profiles (2026)	Model picks realistic watch percentages per video length	Randomized retention that follows real audience curves
AI proxy health scorer (2026)	Scores proxies by failure rate and latency trend	Skips dying proxies before they burn a session

---

🧨 The Problem

• Manual view farms get flagged because every session shares one IP or one UA string.
• Naive bots play 100% of the video every time — retention curves look artificial and views get filtered.
• Proxy pools silently degrade: you only discover a dead proxy after a session fails to register a view.
• Playback timers drift on slow connections, so a “watch 4:30” script actually delivers 3:12 or 6:40 of watch time.
• Long jobs crash mid-run and there’s no resume — you start the whole thing over.
• Fingerprint mismatch (IP in one country, timezone in another) is a top cause of dropped views.
• No visibility into why views didn’t count — you get a number and nothing else.

---

🛠️ The Solution

Problem	Solution
Shared IP/UA across sessions	Per-session proxy binding + locked fingerprint at spawn
Artificial 100% retention	Retention timeline scripting with random pause/seek events
Dead proxies discovered late	AI proxy health scorer pre-filters failing endpoints
Playback timer drift	4-second drift correction against actual currentTime
No crash recovery	SQLite job state + resume-from-last-session
Fingerprint mismatch	Geo-consistent fingerprint assembly (timezone ← proxy region)
Zero failure diagnostics	Per-session log with failure reason (proxy fail, view filtered, playback error)

---

🚀 Quick Start

1. 🧾 Visit the project page and hit the download section — you get a zip containing the .exe and a config/ folder.
2. 📦 Extract the zip to a folder you own (not Program Files — the app writes job state next to itself).
3. ▶️ Run the .exe — first launch creates config/settings.json and an empty proxy pool.
4. 🧩 Load your proxy list in the Proxy Manager tab (one host:port:user:pass per line, protocol dropdown per pool).
5. 🎬 Create a job — paste the target video URL, set view count, sessions, and pick a retention profile. Hit Start.

First run will ask you to accept a one-time notice about YouTube ToS and local data storage. Nothing is sent anywhere except the YouTube requests themselves.

---

🧩 Module Catalog

44 modules across nine subsystems. Status reflects the current 2026 build.

Module	Status	Description
SessionSpawner	✅ Working	Creates isolated browser profiles with locked fingerprint
FingerprintSampler	✅ Working	Captures canvas, WebGL, fonts, UA, timezone per session
GeoResolver	✅ Working	Matches timezone/locale to proxy region
CookieJarManager	✅ Working	Encrypted per-session cookie persistence
ProxyPool	✅ Working	Pool container with health state per endpoint
ProxyRotator	✅ Working	Round-robin / sticky / per-job rotation strategies
ProxyHealthScorer	✅ Working	AI-scored endpoint reliability (2026)
ProxyImportParser	✅ Working	Parses host:port:user:pass and URI formats
RetentionScripter	✅ Working	Builds play/pause/seek timeline from profile
RetentionProfiles	✅ Working	Named retention templates (short, medium, long-form)
DriftCorrector	✅ Working	4s re-sync against actual player currentTime
PlaybackController	✅ Working	Play, pause, seek, mute, skip events
WarmUpWalker	✅ Working	Pre-target homepage/related browsing
ViewCounter	✅ Working	Tracks accepted view deltas from telemetry
ThrottleGuard	✅ Working	Caps concurrent sessions per job
JobScheduler	✅ Working	Queue and run multiple jobs sequentially
JobStore	✅ Working	SQLite persistence of job + session state
ResumeEngine	✅ Working	Resume crashed job from last completed session
HeadlessRunner	✅ Working	CLI batch execution without GUI
CredentialVault	✅ Working	AES-encrypted storage for proxy creds
ConfigManager	✅ Working	settings.json load/validate/migrate
LogRotator	✅ Working	Rotates session logs by size/date
SessionLogger	✅ Working	Per-session event log with failure reason
TelemetryPanel	✅ Working	Live view delta, session health, proxy latency
LatencyMonitor	✅ Working	Per-proxy RTT sampling
FailureClassifier	✅ Working	Tags failures: proxy, playback, filtered, timeout
AiRetentionPicker	✅ Working	Model selects retention % from video length
AiProxyAdvisor	✅ Working	Recommends pool swaps mid-job
FingerprintConsistencyCheck	✅ Working	Flags IP/geo/UA mismatch before spawn
SessionPoolRecycler	✅ Working	Recycles dead or flagged sessions
RateLimiter	✅ Working	Per-proxy request pacing
UAProfileLibrary	✅ Working	Curated UA + viewport pairs per browser family
CookieSeeder	✅ Working	Seeds sessions with plausible prior activity
VideoResolver	✅ Working	Resolves video ID from URL variants / shorts
PlaylistWalker	✅ Working	Handles playlist and related-video targets
DwellTimer	✅ Working	Adds non-playback dwell time on the watch page
MuteRandomizer	✅ Working	Randomized mute/unmute during playback
QualitySwitcher	✅ Working	Random quality selection per session
AdSkipper	✅ Working	Handles pre-roll skip timing
StatsAggregator	✅ Working	Post-job stats: views, filtered, failed, avg watch
ExportModule	✅ Working	Exports job stats to CSV
UpdateChecker	✅ Working	Checks landing page for newer build
DiagnosticsBundle	⚠️ Beta	One-click zip of logs + config for support
ProxyAutoRefill	⚠️ Beta	Pulls replacement proxies from configured provider

---

🕸️ Session & Proxy Management

Session isolation — each session is a fully separate browser profile on disk under sessions/. Cookie jar, fingerprint, and proxy binding are written at spawn and never mutated.

Proxy binding rules

• sticky: one proxy → one session for the job’s lifetime
• round-robin: proxies cycle every N sessions (configurable)
• per-job: one proxy pool per job, isolated from other jobs
• pool-shuffle: reshuffles the pool every 10 sessions

Health scoring — the ProxyHealthScorer tracks failures, timeouts, and latency trend per endpoint. Proxies below the threshold are quarantined rather than deleted, so you can inspect why they failed.

Fingerprint assembly order — proxy region → timezone → locale → UA family → viewport → canvas/WebGL. Each value depends on the previous, which is why you can’t just randomize the UA independently.

---

⏱️ Retention & Watch-Time Engine

The retention engine is what separates a working bot from a view-filtered one. A fixed “watch 100%” pattern is the single most common cause of dropped views.

Timeline format (retention profile)

{
  "name": "medium-random",
  "watch_pct": [0.45, 0.92],
  "events": [
    { "at": 0.12, "action": "pause", "duration": [3, 9] },
    { "at": 0.40, "action": "seek", "delta": [-8, 8] },
    { "at": 0.70, "action": "mute", "duration": [5, 20] }
  ]
}

watch_pct is a range — each session samples a random value inside it. AiRetentionPicker overrides this range per video using length-aware curves (short videos skew higher, long-form skews lower and more variable).

Drift correction — every 4 seconds the engine compares the scripted timeline position to the actual currentTime from the player. If drift exceeds 2.5 seconds, the script re-anchors. Without this, slow proxies push the timeline out of sync and the session “watches” the wrong segment.

---

🖥️ System Requirements

Component	Minimum	Recommended
OS	Windows 10 64-bit (build 1909+)	Windows 11 23H2+
CPU	4 cores / 4 threads	8 cores / 16 threads
RAM	8 GB	16 GB (16+ sessions concurrent)
Disk	2 GB free	8 GB free (session profile cache)
Network	10 Mbps down	50 Mbps down, low jitter
Proxy pool	1 endpoint per concurrent session	5–10 endpoints per concurrent session
Display	1280×720	1920×1080
Admin rights	Not required	Not required
Runtime	None (self-contained)	None

Tips for best results

• Keep concurrent sessions at 3 or below for a single target video — the throttle guard defaults to this for a reason.
• Use residential or mobile proxies for the target video; datacenter proxies are the fastest route to filtered views.
• Warm up sessions (WarmUpWalker) — sessions that jump straight to a target video behave differently than ones that browse first.
• Don’t point two jobs at the same video simultaneously; the view counter aggregates per video and you’ll lose track of which job contributed what.
• Keep at least 5 proxies beyond your concurrency cap so the health scorer has room to quarantine without starving the pool.
• Run headless batch mode from Task Scheduler if you want overnight jobs; the GUI adds ~15% CPU per session.

---

📥 Installation

1. Download the build. Go to the project landing page, download the current zip. The archive contains the .exe, a config/ skeleton, and a short readme-first.txt.
2. Extract to a writable folder. The app writes jobs.db, sessions/, and logs/ next to the executable. Avoid Program Files and any path with non-ASCII characters.
3. Run the .exe. First launch runs a config check, creates defaults, and opens the dashboard. If Windows SmartScreen prompts, choose “More info → Run anyway” — the build is unsigned.
4. Configure proxies. Open the Proxy Manager tab, paste your list, pick the protocol per pool, and hit Validate. Anything red is dead on arrival.
5. Create your first job. Target video URL, view count, sessions, retention profile. Start small (5 views) to confirm the pipeline works end-to-end before scaling up.

---

🧠 AI Integration (2026)

Three AI-assisted subsystems are active in this build. None of them send your data to a remote model — the models are small, local, and shipped inside the .exe.

AiRetentionPicker — Instead of a fixed watch-percentage range, this picks a retention value from a distribution fit against real audience retention curves for the video’s length category. A 3-minute video might sample 70–95%; a 40-minute video samples 18–55% with more variance. This is the single biggest improvement over fixed-range retention.

AiProxyAdvisor — Samples proxy latency and failure rates mid-job, and when a pool’s health drops below threshold it suggests (or auto-applies, if enabled) a pool swap. It also flags proxies that respond fine but consistently fail to register views — a failure mode latency monitoring misses.

FailureClassifier — Post-job, classifies each failed session into proxy failure, playback error, filtered view, or timeout. This replaces the old “count of failures” with a breakdown you can actually act on.

Real-time updates are also new: the telemetry panel now streams view-delta and session health over a local WebSocket instead of polling every 5 seconds, so a dying proxy shows up within ~1 second.

---

🏗️ How It Works

The application is a session-orchestration pipeline. When you start a job, the JobScheduler asks the SessionSpawner for N sessions. The spawner pulls endpoints from the ProxyPool, runs each through FingerprintConsistencyCheck, then builds a browser profile via FingerprintSampler and GeoResolver — timezone and locale are derived from the proxy’s geolocation, not randomized independently. If the consistency check fails (proxy says Germany, fingerprint says Vietnam), the session is rejected before it ever touches YouTube.

Once a session is live, WarmUpWalker optionally browses the homepage and a couple of related videos for 30–90 seconds. Then VideoResolver turns your target URL into a video ID, PlaybackController navigates to the watch page, and RetentionScripter builds the timeline from the selected profile (or from AiRetentionPicker if AI mode is on). PlaybackController executes the timeline, while DriftCorrector samples currentTime every 4 seconds and re-anchors if the script has drifted. DwellTimer, MuteRandomizer, and QualitySwitcher layer on the smaller behavioral details.

ViewCounter reads the view-count delta from the watch page after the session ends, and FailureClassifier tags any session that didn’t produce a delta. Job state is written to SQLite after each completed session by JobStore, which is what makes ResumeEngine work — a crash at session 47 of 200 resumes at session 47, not session 1. StatsAggregator produces the end-of-job summary and ExportModule writes it to CSV if you want the data elsewhere.

---

⚖️ Comparison

Aspect	Generic view bots	Browser-extension “viewers”	yt-view-bot
Session isolation	Shared browser profile	Runs in your real browser	One profile per session, locked
Proxy handling	Manual, one IP	None	Pool with health scoring + rotation
Retention realism	Fixed 100% play	Whatever you do manually	Scripted timeline + AI curve sampling
Drift handling	None	N/A	4s re-sync against currentTime
Crash recovery	Restart from zero	N/A	Resume from last completed session
Failure diagnostics	Count only	N/A	Classified per session
Batch scheduling	Manual	N/A	Headless CLI + Task Scheduler
Fingerprint consistency	Random UA	Your real fingerprint	Geo-consistent assembly

Fair framing: the extension route is genuinely simpler if you just want a handful of views from your own browser — this tool is for people running hundreds of sessions and needing them to look separate. Generic bots are fine for targets that don’t validate hard; this one is built for the ones that do.

---

🧯 Known Issues

Issue	Solution
Views register but get filtered after ~48h	Retention profile too uniform — switch to AiRetentionPicker and raise watch_pct variance
Session hangs on watch page with no playback	Usually a dead proxy that passes latency check but fails media requests; quarantine it manually
DiagnosticsBundle sometimes omits the last session log	Known beta issue — the log flush races the bundle build; wait 5s after job end before bundling
High RAM usage above 12 concurrent sessions	Expected — each session holds a browser profile in memory; drop concurrency or raise RAM
Proxy validation passes but spawn fails	Fingerprint consistency check rejecting geo mismatch; loosen the check or fix proxy region metadata
Headless mode won’t pick up config changes	Config is loaded once at process start; restart the headless process after editing settings.json

---

📜 Changelog

Version	Date	Changes
2026.3	2026-03-18	AiProxyAdvisor auto-apply mode; WebSocket telemetry (replaces polling); FailureClassifier adds timeout category
2026.2	2026-02-09	AiRetentionPicker length-aware curves; ProxyAutoRefill beta; drift threshold lowered to 2.5s
2026.1	2026-01-14	2026 baseline: geo-consistent fingerprint assembly, ResumeEngine hardened, encrypted vault migration
2025.4	2025-11-02	HeadlessRunner CLI flags expanded; ExportModule CSV format
2025.3	2025-09-20	WarmUpWalker added; SessionPoolRecycler stability fixes
2025.2	2025-07-11	Proxy rotation strategies (sticky / round-robin / per-job)
2025.1	2025-05-04	Initial public build — spawner, playback controller, SQLite job store

---

❔ FAQ

1. Will this get my YouTube account banned? The bot doesn’t log into accounts — sessions are anonymous and never authenticate. There’s no account to ban. The risk is to the target video’s view count (views may be filtered), not to any personal account.

2. Why do my views show up then disappear after two days? That’s YouTube’s delayed view validation. The most common causes are uniform retention (every session watching the same percentage) and proxy reuse across sessions. Switch to AiRetentionPicker, raise watch_pct variance, and check your proxy pool for shared endpoints.

3. Do I need residential proxies? For a target video you actually care about, yes. Datacenter proxies work for testing the pipeline but get filtered at a much higher rate. Mobile proxies are the most reliable and the most expensive.

4. How many concurrent sessions should I run? Default is 3. Below 3 is safer but slower; above 6 starts burning proxies faster than the health scorer can replace them. The ThrottleGuard exists to stop you from setting this too high by accident.

5. Does the AI stuff send my data anywhere? No. The models are small, local, and compiled into the .exe. AiRetentionPicker and AiProxyAdvisor run entirely on-device. The only outbound traffic is the YouTube requests themselves.

6. Can I run this on a server without a display? HeadlessRunner handles that — it’s a CLI flag on the same .exe. No GUI is spawned. You’ll want to run it via Task Scheduler or a service wrapper on Windows Server.

7. What happens if the job crashes halfway? JobStore writes state after every completed session, and ResumeEngine picks up from the last one. You lose at most one session’s worth of progress.

8. Why does the .exe trigger SmartScreen? The build is unsigned. Signing certificates for this kind of tool are impractical to maintain. “More info → Run anyway” is the standard path; the binary is self-contained and does not modify system files.

9. Can I target a playlist or a channel instead of one video? PlaylistWalker handles playlists and related-video targets. Channel-wide targeting is not supported in this build — jobs are scoped to a single video or playlist.

10. Where are my logs and job data stored? Next to the .exe: jobs.db for job state, logs/ for per-session logs, sessions/ for browser profile caches. Nothing is written outside that folder, and the credential vault is AES-encrypted at rest.

---

💬 Community & Support

• Bug reports — open an issue with the DiagnosticsBundle zip attached (wait 5s after job end, see Known Issues).
• Proxy provider discussion — there’s a pinned thread in Discussions for comparing residential/mobile providers by failure rate; keep it factual.
• Feature requests — tag with enhancement; the module catalog above is the current scope, so requests that extend an existing module get prioritized over new subsystems.
• Build updates — UpdateChecker polls the landing page; manual checks are in the About tab.

---

The 2026 build is the first where the AI subsystems actually change outcomes rather than just adding a checkbox — AiRetentionPicker in particular moves the needle on filtered views more than any proxy upgrade. If you’re coming from an older build, the encrypted vault migration runs automatically on first launch, but back up jobs.db first. The module catalog above is the honest scope: 41 working modules, 2 in beta, and everything else is a wishlist item, not a shipped feature.