Agentic Streams: AI Agents That Replace Your Feeds With Video Briefings

By Prahlad Menon 3 min read

The internet’s content delivery model is fundamentally broken. Algorithmic feeds decide what you see, how it’s filtered, and how it’s presented. You’re not in control — the platform is.

Agentic Streams from the VideoDB team flips this model entirely: autonomous AI agents crawl the internet, gather real assets, filter noise and bias, then compile and stream polished video briefings directly to you.

No feeds. No scrolling. No algorithmic manipulation. Just a topic and a finished video.

How It Works

The framework follows a clean five-stage pipeline that any agent can implement:

  1. Research — Deep web crawling via browser-use across news, social media, and video sources (including YouTube). No hallucinated data — everything comes from real sources.
  2. Asset Gathering — Agents collect YouTube clips, tweets, article screenshots, charts, and other multimedia evidence.
  3. Script Generation — Professional voiceover scripts are generated with TTS.
  4. Video Assembly — Multi-track composition using VideoDB’s infrastructure.
  5. Delivery — A stream URL ready to play and share.

Input: a topic. Output: a broadcast-quality video stream.

Available Agents

The repo ships with three production-ready agents:

  • Content Creator — Research any topic and produce a 1-2 minute video briefing. It then self-reviews using VideoDB’s “See + Understand” capability to verify visual-narration alignment frame by frame.
  • News Digest — Multi-source evidence gathering (YouTube, tweets, articles) compiled into a 3-4 minute broadcast-style video report.
  • Financial Market Analysis — Investigates financial markets with real charts, screenshots, and verified clips for custom-length market reports.

Each agent is self-contained, reproducible, and fully autonomous. Give it a topic, walk away, come back to a finished video.

Why This Matters

We’ve seen text-based AI research agents (Deep Research from Google and OpenAI, Perplexity, etc.), but video as the output medium is a different game entirely. Video briefings are:

  • Higher bandwidth — Charts, clips, and narration convey more than bullet points
  • Evidence-based — Every claim is sourced with actual multimedia assets, not AI-generated imagery
  • Consumable — Watch a 3-minute briefing instead of reading 15 articles

The self-review loop in the Content Creator agent is particularly interesting. It uses VideoDB’s vision capabilities to verify that what you see actually matches what the narration says — a built-in hallucination check for video content.

The Bigger Picture

This sits at the intersection of two trends: agentic AI workflows and the growing backlash against algorithmic content curation. Instead of platforms deciding what content reaches you, you tell an agent what you care about, and it does the journalism for you.

The Reddit discussion around this project highlights the appetite for this kind of tool — people are tired of feeds and want agent-curated content on their own terms.

Getting Started

Setup is straightforward:

npx skills add video-db/skills
export VIDEO_DB_API_KEY=your_key_here

Then give any agent a topic:

Create a video report about "climate summit 2026"

You can also build your own agents by following the existing folder structure (SKILL.md, AGENTS.md, README.md per agent).

The framework is built on VideoDB, which provides the underlying infrastructure for agents to see, hear, and understand multimedia content. A free API key is available at console.videodb.io.

Final Thoughts

Agentic Streams represents a compelling vision: what if your news, market updates, and research came as polished video briefings assembled by agents that work for you, not for an ad platform? The execution is solid, the agents are production-ready, and the self-review capability adds a layer of quality assurance that most AI content pipelines lack.

Worth watching — both the project and the videos it produces.

Links: