Open source · Apache-2.0 · v0.1 released

Turn any document into a detailed explainer video.

Vyakhya is an open-source, multi-agent engine that reads a PDF and produces a deep, accurate, fully editable explainer video — and it runs entirely on your own machine.

No account No cloud Bring your own models

↓ scroll to watch

vyakhya · attention-is-all-you-need.pdf → out.mp4

Watch a document become a video · 2:14

02:14

The problem

Knowledge is trapped in dense PDFs. The tools meant to unlock it are shallow, closed, and out of your hands.

Closed black boxes

Doc-to-video tools guard their pipelines. You can't inspect, extend, or trust the output.

Shallow summaries

Most tools skim the abstract and produce a 90-second reel. Figures, equations, and nuance get lost.

No real editing

Once rendered, the video is an opaque MP4. You can't tweak a scene, fix a claim, or re-render.

How it works

A crew of agents, not a single prompt.

Vyakhya orchestrates specialized agents — each with a narrow job — so the final video is deep, grounded, and inspectable at every step.

01
Read PDF
Parse text, figures, equations.
02
Comprehend
Build a semantic map of the document.
03
Plan
Outline scenes and narrative arc.
04
Script
Draft narration per section.
05
Design visuals
Generate HTML-native scenes.
06
Narrate
Voice each scene with your TTS.
07
Verify
Cross-check every claim against sources.
accuracy layer
08
Assemble
Stitch scenes, timing, transitions.
09
MP4
Deterministic, editable output.

The verifier is the difference. Every claim in the narration is checked against the source PDF and annotated with a citation. Hallucinations don't make it into the render.

Features

Built for depth, not shallow summaries.

Multi-agent depth

Section by section, figure by figure, equation by equation — not a 90-second summary.

Fully editable

The video is structured data. Edit scenes, rewrite narration, tweak visuals, re-render.

Runs on your machine

Self-hosted. Your PDFs, keys, and outputs never leave your hardware.

Bring your own models

OpenAI, Anthropic, or fully local via Ollama. Your keys, your choice, your cost.

Accurate & grounded

A verifier agent cross-checks every claim against the source and cites it inline.

No limits

Render as many videos, as long, as often as you want. No credits, no quotas.

Open source

Apache-2.0. Read the code, fork it, extend it, ship it inside your own tools.

HTML-native rendering

Scenes are code — deterministic, diffable, version-controllable, reproducible.

Editable, not a black box

Meet the studio.

The output isn't an opaque MP4 — it's a structured project. Open Vyakhya Studio, tweak narration, swap visual types, adjust parameters, then re-render deterministically.

vyakhya studio · /projects/attn-2024/editor

Attention Is All You Need — Deep Dive

Editable explainer · 5 scenes · 16:9

Scene 02 · diagram.attention
Q
K
V
softmax(QKᵀ / √dk) · V
Live preview (HTML)0:12 / 0:30
Scene list you can reorder, split, and delete
Inline narration editor with citation pins
Per-visual parameters — no prompt guessing
Deterministic re-render — no surprises

Self-host

Clone, run one script, open your browser.

No account. No cloud. No email required. Vyakhya is free forever and runs entirely on your hardware.

1
Clone
One repo, one script. Docker handles the rest.
2
Configure models
Point to OpenAI, Anthropic, or your local Ollama.
3
Generate
Drop a PDF, hit render, edit, ship.
Docker
Postgres
React
Python
FFmpeg
Ollama
~/dev/vyakhya
# Get Vyakhya running in under 2 minutes
$ git clone https://github.com/Shrijeeth/Vyakhya.git
$ cd vyakhya && ./setup.sh
$ open http://localhost:3000

# Drop a PDF, pick your models, hit render.

Free · No account · No cloud · Your data never leaves your machine

Open source

Built in the open. Powered by contributors like you.

Vyakhya is Apache-2.0 licensed and community-driven. Star the repo, file an issue, open a PR — every contribution shapes the roadmap.

0
Stars
0
Forks
0
Contributors
0
Discussions

github.com/Shrijeeth/Vyakhya · Apache-2.0

FAQ

Straight answers.

Still curious? The full docs and community discussions have more.