Socials

AI-powered social media content engine. Clip videos, generate posts, publish everywhere, and learn what works — all through natural conversation.

Built by Soshilabs

What it does

• Video clipping — Drop a YouTube URL or upload a .mov/.mp4. AI finds viral moments, cuts short-form clips, adds captions, and formats for TikTok/Reels/Shorts.
• AI drafting — Generate tweets, threads, LinkedIn posts, Instagram captions. Swipe through them Tinder-style or review in a list.
• Multi-platform publishing — Publish to 14+ platforms via Zernio: TikTok, YouTube, Instagram, Facebook, X, LinkedIn, Pinterest, Reddit, Bluesky, Threads, Telegram, Snapchat, WhatsApp, Google Business.
• AutoScore — Self-improving feedback loop. Compares predicted virality scores against real engagement data and adjusts scoring weights over time.
• Content autopilot — Fetches trending topics from tracked creators, analyzes what performs well, generates draft posts daily.
• Brand profile — AI learns your tone, audience, and content pillars from your website. All generated content matches your voice.
• Real analytics — Pull engagement data from Zernio (views, likes, comments, shares, engagement rate) with platform breakdowns.
• MCP server — 57 tools exposed over JSON-RPC for integration with Claude Desktop, Soshi Desktop, or any MCP client.

Stack

Layer	Tech
UI	Next.js 16, React 19, Tailwind CSS 4, Framer Motion
AI	Vercel AI SDK, Claude/Gemini models
Publishing	Zernio API (OAuth connect + post management)
Auth	BetterAuth + Convex
Pipeline	FFmpeg, yt-dlp, Cobalt
Data	JSON file stores (local), Convex (cloud)

Getting started

# Clone
git clone https://github.com/dylanworrall/clipbot.git
cd clipbot

# Install dependencies
npm install
cd ui && npm install

# Set up environment
cp .env.example .env
# Add your keys:
#   GOOGLE_GENERATIVE_AI_API_KEY=...
#   LATE_API_KEY=sk_...  (Zernio API key)

# Run the UI
cd ui && npm run dev

Open http://localhost:3000

Environment variables

Variable	Required	Description
GOOGLE_GENERATIVE_AI_API_KEY	Yes	Gemini API key for AI analysis
LATE_API_KEY	Yes	Zernio API key for publishing + analytics
COBALT_URL	No	Self-hosted Cobalt instance for downloads (default: http://localhost:9000)
CONVEX_DEPLOYMENT	Cloud only	Convex deployment for auth + cloud storage
NEXT_PUBLIC_CONVEX_URL	Cloud only	Convex URL (enables cloud mode)

Features

Video pipeline

Paste a YouTube URL or drag-and-drop a video file. The pipeline:

1. Downloads the video (Cobalt or yt-dlp)
2. Transcribes with word-level timestamps
3. AI analyzes transcript for viral moments (scored 1-10)
4. FFmpeg cuts clips with background fill + captions
5. Clips appear in the UI ready for review and publishing

Supports: YouTube, Twitch, Kick, TikTok, Instagram, X, Facebook, Reddit, Dailymotion, Vimeo. Plus local .mov, .mp4, .mkv, .avi, .webm files.

AI chat

Conversational interface with 30+ tools. Examples:

• "Process this YouTube video: [url]"
• "Generate 5 tweets about AI productivity"
• "Run the autopilot"
• "Connect my TikTok account"
• "Show my analytics for the last 30 days"
• "Fill my queue with linkedin posts and threads"

Social connectors

Connect accounts via Zernio OAuth. Multiple accounts per platform supported (each gets its own Zernio profile). Publishing targets all connected accounts automatically.

AutoScore

Learns from real post analytics to improve virality predictions:

• Collects engagement data from all published posts
• Normalizes to 1-10 scale via percentile ranking
• Pearson correlation tracks prediction accuracy
• Per-category weight adjustments (education, entertainment, controversy, etc.)
• Decay-weighted for recency — recent data matters more

Content queue

AI generates content in multiple formats (tweet, thread, LinkedIn, caption, script, meme). Review in:

• List view — Edit inline, publish, or delete
• Swipe view — Tinder-style cards, swipe right to approve, left to reject

MCP server

JSON-RPC endpoint at /api/mcp exposes all tools for external AI agents:

curl -X POST http://localhost:3000/api/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Project structure

clipbot/
  src/                  # CLI pipeline (TypeScript)
    modules/            # downloader, transcript, analyzer, clipper, publisher, captions
    pipeline/           # runner, state management
  ui/                   # Next.js web app
    src/
      app/              # Pages (chat, drafts, calendar, analytics, creators, settings)
      components/       # UI components (Soshi dark design system)
      lib/
        ai/             # AI tools (tools.ts), system prompt, tool definitions
        mcp/            # MCP server (handler, auth, crypto, rate limiting)
        *.ts            # Stores (brand, queue, schedule, autoscore, settings, creators)
      hooks/            # React hooks (useSettings, useAiChat, useThreads)
    convex/             # Convex schema + functions (cloud mode)
  dist/                 # Compiled CLI + MCP tools

Settings

Accessible at /settings with tabs:

Tab	What it does
General	Account, API keys, Claude model config
Brand	AI-powered brand profile (scrape URL, edit tone/audience/topics)
Style	Pipeline defaults, caption styling, background fill
Scoring	Virality weight sliders (hook, standalone, controversy, education, etc.)
AutoScore	Self-improving feedback loop config + learning cycle
Autopilot	Daily content generation config (posts/day, preferred time, platforms)
Connectors	Social account connections (14 platforms via Zernio)
Billing	Subscription management (cloud mode, via Whop)

License

Private. Copyright Soshilabs 2026.