Understand your creator persona across Douyin, TikTok, and Red Note — with zero AI.

_{Click to watch on YouTube →}

Try the live demo →

Content creators manage multiple platforms but have no unified view of their performance. Each platform's analytics lives in a silo, metrics aren't comparable, and most "analytics tools" are either expensive SaaS products or black-box AI that can't explain its recommendations.

DashPersona ingests your creator data from Douyin, TikTok, and Red Note, normalizes it into a unified schema, and runs it through 11 deterministic analysis engines backed by an 18-signal intelligence framework. Every score, tag, and recommendation is computed with statistically validated algorithms — Hazen percentile ranking, OLS regression with significance testing, adaptive sample-size thresholds, and platform-specific signal weighting. No LLM calls, no API keys, no subscription fees. You can trace any number back to the formula that produced it.

Visit dash-persona.vercel.app and click Try Demo to explore with built-in sample data. No login, no setup. You can also upload your own XLSX/JSON exports from Creator Centers to see real analysis — but the web version is limited to file-based data.

Download the DASH Collector desktop app (macOS / Windows). It runs in your system tray. On first launch, log in to your Creator Center inside the built-in Chromium browser — after that, the DASH web app automatically collects data through the Collector.

• No CLI tools required
• No programming knowledge needed
• Persistent cookies — no need to re-login after restart
• System tray shows collection status

Download: GitHub Releases (coming soon)

Install the CLI for automatic data collection directly from your Chrome browser. Supports Douyin, TikTok, and Red Note creator centers using your existing login sessions.

npm install -g @anthropic-ai/claude-code
claude skill install --global github.com/eze-is/web-access

New to the command line? Follow the step-by-step installation guide — designed for complete beginners with no programming experience.

Growth sparklines, cross-platform follower deltas, niche benchmarking, and strategy suggestions — all on one screen.

A composite 0–100 score breaking down your content mix, engagement rate, posting rhythm, persona consistency, growth health, and viral potential. Click any dimension to see the exact formula.

Side-by-side metrics across all your platforms. Automatically surfaces insights like "your audience on Douyin is 4.2x larger than on Red Note" and "your content gets 1.8x more engagement on Red Note than TikTok."

git clone https://github.com/Fearvox/dash-persona.git
cd dash-persona
npm install
npm run dev

Open localhost:3000 and click Try Demo to explore with sample data.

For real data collection, see the installation guide.

  Your Data                    Analysis Pipeline                   What You See
 ───────────                  ──────────────────                  ──────────────
 CDP proxy (Douyin,           ┌─ stats/ library ──────────┐       Dashboard
   TikTok, XHS) ──┐          │  (percentile, regression, │       Persona Detail
 JSON / CSV ───────┤          │   normalize, threshold)   │       Compare View + Radar
 XLSX (11 schemas) ┼─ Quality ┼─ Persona Score            │       Content Calendar
 Chrome extension ─┤  Filter  ├─ Growth Tracker           ├──┐    Persona Timeline
 Manual import ────┘    │     ├─ Niche Detection          │  │    Data Portrait
                    Profile   ├─ Niche Benchmark          │  │    Pipeline View
                    Store     ├─ Strategy Engine           │  │    Export (PNG/PDF/CSV)
                  (IndexedDB) ├─ Content Planner           │  │
                              ├─ Cross-Platform Comparator │  │
                              ├─ Idea Generator            │  │
                              ├─ Content Analyzer          │  │
                              ├─ Next Content Engine       │  │
                              └─ Signal Collector (18) ────┘  │
                                 Platform weights:            │
                                 Douyin: completion rate 8x   │
                                 XHS: save rate 6x            │
                                 TikTok: viral ratio 7x ──────┘

All engines are deterministic. Same input always produces the same output. No randomness, no model weights, no external API calls.

• Signal framework — 18 standardized signals across 5 categories (engagement, rhythm, growth, content, audience) with platform-specific weights derived from Twitter/X UUA patterns
• Statistical foundation — stats/ library with Hazen plotting-position percentiles, OLS linear regression with t-test p-values, adaptive sample-size thresholds, and rank-based normalization
• Content classification — inverted-index keyword matching across 31 categories with cached results
• Engagement scoring — weighted formula with post quality filtering (short text, zero engagement, and bot activity down-weighting)
• Persona consistency — adaptive sliding-window cosine similarity (window auto-scales: 5 for 30 posts, 20 for 200 posts)
• Trend analysis — OLS regression replaces naive difference-of-means, with trendReliable flag gated by p < 0.05
• Niche detection — maps content distribution to 10 benchmark niches with Hazen percentile ranking (no tail collapse)
• Growth analysis — delta computation over IndexedDB-persisted historical snapshots with engagement velocity signal
• Engine memoization — FNV-1a content hashing with LRU eviction (maxSize=64)
• XLSX schema detection — auto-classifies 11 export formats (4 Douyin + 7 TikTok Studio) with cross-year date boundary fix
• Profile persistence — IndexedDB storage with sessionStorage sync cache, merge-on-import for cross-session data retention
• Algorithm optimization — pre-computed keyword overlaps (-75% rule execution ops), cached category lookups, deterministic benchmark data

src/lib/engine/         — 11 analysis engines + stats/ library + signal-collector (18 signals)
src/lib/engine/stats/   — statistical primitives (percentile, regression, normalize, threshold)
src/lib/adapters/       — 7 data adapters (CDP, Demo, FileImport, Manual, Extension, HTMLParse, Browser)
src/lib/store/          — IndexedDB profile persistence with sessionStorage sync
src/lib/collectors/     — CDP client, trending collector, video analyzer, temp storage
src/app/api/            — CDP collection + trending endpoints
src/app/install/        — Beginner-friendly CLI setup guide
src/app/onboarding/     — Multi-mode data import wizard (file upload, URL paste, CDP auto-collect)

Fallback adapters: HTMLParseAdapter and BrowserAdapter are retained as fallback options. Use CDPAdapter for reliable platform data collection.

China's creator economy exceeds 1.5 trillion RMB annually, but individual creators still fly blind. Platform analytics are siloed, metrics aren't comparable, and third-party tools charge monthly subscriptions for black-box outputs. DashPersona gives every creator a transparent, self-hosted intelligence engine — at zero cost.

DashPersona was built across 20+ sessions using EverMemOS as the persistent memory layer for AI-assisted development. EverMemOS is an open-source structured memory system for AI coding agents — it automatically captures architectural decisions, sprint plans, and debugging context into a queryable knowledge base.

This project is a living case study: 11 analysis engines, 7 data adapters, a Chrome extension, and a cinematic landing page — all designed across separate sessions but maintaining consistent interfaces. Architecture decisions made in Session 1 were correctly recalled and enforced in Session 20. Every non-obvious choice is traceable through structured memory.

11 analysis engines. 18 intelligence signals. 7 data adapters. 308 tests. 1 Chrome extension. Statistically validated. Zero AI.

• Niche-aware benchmarking (10 niches, synthetic cohort comparison)
• Browser extension for one-click Douyin data capture
• Client-side history persistence (IndexedDB)
• Platform-specific quality signals (completion rate, bounce rate, watch duration)
• Multi-file import with 4 Douyin XLSX schema auto-detection
• Radar chart multi-dimensional cross-platform comparison
• Report export as PNG and PDF
• Engine memoization with FNV-1a content hashing and LRU eviction
• E2E test coverage with Playwright (15 test cases across 5 core flows)
• Accessibility: focus-visible, skip-to-content, semantic landmarks, keyboard navigation
• CDP-based data collection for Douyin, TikTok, and XHS
• DASH Desktop Collector (Electron app, system tray, persistent Chromium)
• Trending analysis with real-time topic collection
• Next Content Engine with 7 deterministic suggestion rules
• Content structure analysis with video frame extraction
• Enhanced onboarding wizard with CDP proxy setup and platform login verification
• Temp data management with automatic 15-day expiry and 500MB cap
• CSV export for dashboard data
• Analysis delta tracking (change detection between analysis runs)
• TikTok Studio XLSX import (7 additional schema types with date normalization)
• Persistent profile storage (IndexedDB with merge-on-import)
• Demo/real data separation across all detail pages
• CLI installation guide for non-technical users
• i18n support (Chinese/English, 700+ keys, pipeline visualization, detail pages)
• Textcraft ASCII art engine with brand typography and data portrait
• Statistical foundation: Hazen percentile, OLS regression, adaptive thresholds
• 18-signal intelligence framework with platform-specific weights
• Post quality scoring (bot detection, low-effort filtering)
• Persistent navigation with SiteHeader across all pages
• Portrait page with demo mode and CJK-aware column alignment
• Signal dashboard with 18-signal radar visualization
• Audience demographic insights (fanPortrait integration)
• Two-stage ranking pipeline (candidate generation + scoring)

See CHANGELOG.md for a detailed version history.

1. Fork the repo and create a feature branch from main
2. npm install → npm run dev
3. Make your changes, ensure npm run build passes
4. Open a PR with a clear description

Business Source License 1.1 (BSL 1.1)

• Source available — read, fork, and modify freely
• Non-production use permitted without a license
• Production use requires a commercial license from Fearvox
• Converts to Apache 2.0 on 2030-03-24

See LICENSE for the full text.