Product 8 min read

Meet Prism: Ask YouTube anything.

Prism turns a sentence about what you want to watch into a short, hand-picked set of videos — and keeps the personalization on your own machine.

By BeezGroup Product Chrome 116+

YouTube search is built for finding a video you already have in mind. Prism is built for the other half of the problem: expressing what you actually mean, then getting a small, thoughtful set of results that explain themselves.

The gap Prism closes

A search box expects keywords. People arrive with intentions — a mood, an era, a format, a duration, a level of depth. “Something long to listen to while I cook” is not a keyword. Neither is “take me back to 2008,” or “teach me programming from the beginning.”

Prism treats those requests as structured intent. It reads the topic, format, era, duration, level, and mood behind a sentence, then searches YouTube with the user's own session and returns a short deck of cards. Each card includes a one-line reason why it was chosen.

“Take me back to 2008.”

An era, not a keyword — and a different way to navigate a video library.

What Prism does

  • Intent-first discovery. Ask for a mood, an era, a topic, a duration, or a learning goal in plain language.
  • Time machine. Build a keepsake deck of videos from a chosen year across the topics you care about.
  • Learning paths. Turn “I want to learn X” into an ordered, time-budgeted sequence with progress tracking.
  • Answers from inside videos. Summaries, key chapters, and timestamped moments come from the video's own captions.
  • A player that stays with the conversation. Play, queue, minimize, and keep chatting without hopping tabs.

A five-module engine, built on BeezGroup research

Prism is a working implementation of the ideas in BeezGroup's 2022 design document, “Sleeping-Zero Algorithm and Neural Synthesizer.” The architecture is deliberately transparent: five modules, plain rules, and no dependency on a cloud model.

  1. Listening. A request is recorded as a message and a memory, with a small weight bump for detected topics.
  2. Reading. The intent parser extracts topic, format, era, duration, level, and mood.
  3. Synthesizing. The planner builds a YouTube query and collects candidates using the user's session.
  4. Processing. Candidates are scored for affinity, era, duration, format, level, and viewing history.
  5. Delivering. The top picks become cards with reasons, alternative angles, and follow-up options.

Personalization that stays personal

Prism has no account, no cloud, and no telemetry. Its interest graph — the weighted map of topics, channels, eras, and formats that shapes recommendations — lives in Chrome's local storage. The only network traffic is to YouTube itself.

  • No sign-up and no server-side profile.
  • Nothing is collected before onboarding consent.
  • Watch signals only count after 15 seconds of real playback.
  • Data can be exported as JSON or erased completely.
  • Private mode answers a request and forgets it, without touching the taste graph.

What makes it different

Prism is not a general-purpose AI assistant, and it does not pretend to be one. It is a deterministic, explainable rules engine with a curated vocabulary, a fixed set of search behaviours, and a local memory model that slowly forgets. The result is a tool that can say why it recommended something — and can be audited by anyone willing to read the source.

Prism is in its post-development phase. Early-access members get the packaged Chrome build first, along with fresh-picks updates and a direct line for feedback.

Join early access

Availability

Prism is designed for Chrome 116 or newer and is currently free, with no account required. The project is being prepared for release; the full source can already be installed unpacked for developers who want to explore it today.

To see the product site, feature demos, privacy details, and the early-access form, visit prismbrain.netlify.app.