Skip to main content
TikRSS
TikRSS Distillation Engine v0.4

From 2 hours of video,
to 5 minutes of structured signal.

Built for researchers, founders, and thinkers who value attention over algorithmic noise. Capture transcripts, filter sponsored fluff, and wake up to executive syntheses in your knowledge base.

Act I · Capture Everywhere

Capture in 3 seconds without altering your reading habits

Whether you spot a 2-hour podcast on your commute or a technical teardown at your desk, capture it with zero friction:

  • Telegram Bot (@TikRssBot):Share any link directly on mobile to subscribe in 1 tap.
  • 1-Click Bookmarklet:1-click browser button for instant tracking on video pages.
  • Email Forwarding:Forward newsletter digests to your secret inbox address.
  • Direct URL Paste:Paste any YouTube handle, TikTok profile, or open Podcast RSS.
  • OPML / CSV Batch Import:Bulk import YouTube subscriptions or Inoreader OPML in 1 click.
  • Terminal CLI & MCP:Command line tikrss add or AI agent autonomous subscription.
tikrss-ingest-worker
$ tikrss add https://youtube.com/watch?v=lex_karpathy_cot
✓ Resolving author: @AndrejKarpathy (YouTube Verified)
↳ Audio duration: 2h 14m 38s · 144Kbps stream extracted
↳ Queue status: Scheduled for 07:00 morning synthesis
✓ Ready: Full transcript captured & formatted
Act II · Noise Filter

Full Transcripts & Intelligent Sponsor Filter

Beyond raw text, TikRSS surgically detects and excises sponsored ad reads, sponsor transitions, and filler.

AUDIO TRACK TIMELINE · 48:201 SPONSOR SEGMENT REMOVED
Topic Introduction [00:00 - 12:40]
Sponsor -3m15s
Core Argument & Architecture Deep Dive [15:55 - 48:20]
00:0012:40 [VPN / Mattress Ad Stripped] 15:5548:20
99.4%
Transcript Accuracy
-18%
Avg. Fluff & Ads Removed
100%
Timestamp Verifiable
SYNTHESIS07:00 UTC
✓ 5 min read

How model scaling shifts engineering focus toward inference reasoning

Step-by-step token generation explores higher-probability latent verification branches.

Key Takeaways & Citations
  • •Inference compute scaling is overtaking static parameter scaling.
  • •RLHF rewards prose fluency; structured prompting unlocks logical proof.
Why it matters: Shifts how engineering teams allocate AI inference budgets.
Act III · Executive Synthesis

Not generic summaries, but rigorous intelligence briefings

Generic AI summarizers produce vague bullet points. TikRSS structures every insight with rigorous academic and research methodology:

01
Executive TL;DR

Scan the primary breakthrough in 30 seconds.

02
Claims & Empirical Evidence

Anchored directly to audio timestamps to prevent hallucination.

03
Actionable Implication & Paper Citations

Identifies direct industry impact and auto-extracts cited ArXiv papers.

04
Domain-Specific Prompt Templates

Customize focus per channel: Finance metrics, AI compute proofs, or Biotech markers.

05
Priority Processing Queue

Dedicated compute pipeline for Power subscribers with priority morning processing.

Act IV · Second Brain Sync

07:00 AM Delivery & Native Knowledge Base Sync

Your morning briefings arrive before your first cup of coffee, exported directly into your existing productivity toolchain.

07:00 Local Morning Digest

Delivered in your local timezone so you start the day informed without opening social media.

Obsidian Local Vault Sync

Automatic Markdown files with tags, structured metadata, and bidirectional links to your vault.

Notion Database Auto-Entry

Rich page entries with creator tags, cover image, and multi-view database schemas.

Private Encrypted RSS

Tokenized feeds for Reeder, NetNewsWire, and podcast players.

Act V · AI Agent & Terminal Layer

Claude Desktop, Cursor & Terminal Native Toolchain

The first video intelligence engine with official Model Context Protocol (MCP) server support. Give your AI agents direct tools to query recent briefings and verify video facts.

Official MCP Server (5 Agent Tools)

Exposes clean tools for querying today's digest, searching claims with timestamps, and full transcripts.

Terminal Native CLI (tikrss)

Manage feeds, export Markdown, and run full-text query pipelines right from your terminal.

Open Source on GitHub (MIT License)

MIT-licensed open source package ready for local inspection and contributions.

claude_desktop_config.json
MCP Protocol v1.0
{
  "mcpServers": {
    "tikrss": {
      "command": "npx",
      "args": ["-y", "@tikrss/mcp-server"],
      "env": {
        "TIKRSS_API_KEY": "tr_live_••••••••"
      }
    }
  }
}
User: "提炼下 Karpathy 昨晚视频的核心论点"
↳ Agent calls: tikrss_search_claims(query: "Karpathy CoT")
↳ Returns: 3 verifiable claims with timestamps [14:22], [32:10], and paper citations.