Open source · Powered by Claude

Turn any topic into a narrated animation.

Chalkboard is a pipeline of agents that writes a script, verifies it, generates Manim animation code, synthesizes a voiceover, and renders everything to a final .mp4. Fully automated.

Get started → See examples View source ↗

MIT Python 3.10+ · No vendor lock in · Runs locally

Rendered output · 1920×1080 · chalkboard theme

sample · binary search

Binary search — flip to the middle

CC 0:00 / 0:12

The pipeline

Six agents. One video.

Each stage runs as a separate Claude agent with retry logic. Failures in one step don't cascade — they loop and correct.

01

◌

Init

02

✎

Script

03

✓

Verify

04

</>

Animate

05

◳

Visual QA

06

▶

Render

median run · 3m 42s · medium effort retries allowed · up to 3 per stage parallel safe · sandboxed via Docker

What it does

Not just a wrapper. A whole studio.

Everything you'd normally paste prompts into four different tools for — handled in one repeatable run.

01

stage · verify

Every claim gets checked before it gets rendered.

A dedicated verifier agent reads the draft script, annotates numerical claims, dates, and causal statements, and rewrites anything that won't hold up. Only the clean version proceeds.

fact_check.jsonl

"Binary search runs in O(log n) time."

cite · CLRS ch. 2 · verified

"Invented by John von Neumann in 1946."

disputed · first published 1946 by Mauchly; rewriting…

"Requires the input array to be sorted."

cite · Knuth TAOCP vol. 3 · verified

02

stage · animate

Manim scene code, written for you.

The generator reads the approved script beat by beat and emits valid Manim, positioned, timed, and tagged with chapters. Code that won't compile bounces back for a revision before a frame is rendered.

scene_binary_search.py

class BinarySearch(Scene):
    def construct(self):
        arr = [3, 7, 12, 19, 25, 31, 47]
        boxes = VGroup(*[Square() for _ in arr])
        boxes.arrange(RIGHT, buff=0.15)
        # chapter: introduce the array
        self.play(LaggedStart(*[Write(b) for b in boxes]))

03

stage · visual qa

Claude watches the render, not just the code.

After rendering, frames are sampled at fixed intervals. Claude flags overlapping text, elements that run off the canvas, unreadable contrast, and obscured labels, then dispatches another render with targeted fixes.

qa_report · 6 frames sampled

PASS t=0:04

FLAG t=0:11

PASS t=0:17

1 flag · reshooting frame 011 with offset Y +0.4 · eta 14s

04

stage · finalize

Captions and chapters, ready on day one.

Every render produces an .srt, embedded chapter atoms, and a pasteable YouTube chapter list. No extra post production pass.

binary-search.srt · 3 of 7

0:00 Here's a sorted list of seven numbers. Ch 1 · Setup

0:12 Binary search works like a dictionary — flip to the middle. Ch 2 · Idea

0:28 Each step cuts the search space in half. Ch 3 · Analysis

Configurable

Tune the script before it's a video.

Pick an audience, a tone, a template, and a theme. Your prompt doesn't change, but Chalkboard retargets the output.

Audience

Tone

Template

Theme

TTS backend

Effort

script preview · binary search · beginner · casual

est. 1m 20s · 160 wpm render · ~3m 40s

Made with Chalkboard

What the final cut looks like.

Short explainer videos, rendered locally. Each has captions, chapters, and a full source script on disk.

1:26

▶

How the Internet Actually Works

medium · chalkboard

0:55

▶

How Hash Tables Actually Work

medium · chalkboard

1:02

▶

SQL vs NoSQL Databases

medium · compare

0:50

▶

How HTTPS Keeps Your Data Safe

medium · howto

Get started

Three ways in. Pick yours.

Chalkboard is in closed beta. Run it yourself today, sign in with an invite code, or drop your name on the waitlist.

MIT License

01

Run it yourself.

Clone the repo, add your Claude key, run the server.

01 Clone & install

Requires Python 3.10+, Docker, ffmpeg.

$ git clone https://github.com/nicglazkov/Chalkboard.git
$ cd Chalkboard
$ pip install -r requirements.txt

02 Add API keys

$ cp .env.example .env
# then edit .env:
ANTHROPIC_API_KEY=sk-ant-…
TTS_BACKEND=openai
OPENAI_API_KEY=sk-…

03 Run

# web UI — localhost:8000
$ python run_server.py

# or one-shot from terminal
$ python main.py --topic "B-trees" --effort medium

Clone on GitHub ↗

Full setup guide →

Fastest

02

Join the closed beta.

Sign in with an invite code and start making videos right away — we've set everything up for you.

Free while in beta. Invite code required, everything else is handled for you.

Sign in →

New here? Create account →

New invites weekly

03

No code? Get notified.

Invites open in batches. We'll email when yours does.

Request access →