botstop 0.3.0

Anti-AI captcha: moving static hides a code humans read over time, not from a screenshot.

BotStop

Stop bots, not humans.

Live demo — try the captcha in your browser.

BotStop is an anti-AI captcha where the answer lives in motion, not in pixels. The digits are cut from the same TV static as the background and bounced across the field. Pause on any frame and there is nothing to read — freeze time, and all discernment is lost to noise. Humans track the moving patch instinctively; bots need a persistent model that understands video, not a screenshot and an OCR pipeline.

Why BotStop

Most captchas make the image harder: warp text, add clutter, distort glyphs. Bots respond with better OCR. BotStop sidesteps that arms race by changing the modality entirely.

Typical captcha	BotStop
Hides text under noise	Text is the noise — same field, cut and moved
Escalates distortion to beat OCR	No stable glyph edges to latch onto
Solvable from a single frame with enough ML	A single frame gives almost nothing
More complexity on a static image	Watch the sequence; can't screenshot it

When the challenge becomes a static image instead of a process, the signal disappears. That is the design.

How it works

1. Generate a full canvas of point static — a fresh random field for every challenge.
2. Cut the answer from that same field. Not overlaid, not filtered: the same pixels.
3. Move the cutout across the base static (DVD-style bounce).
4. Verify that the user watched and typed what they saw.

Each challenge gets a new static field, answer, and motion path. Nothing carries over between instances, so there is no shared background for reinforcement learning to memorise. Within a single challenge, the base static stays fixed; the security property is temporal integration over time, not pattern matching across sessions.

Security model

BotStop does not aim for perfect security — that would break legitimate use. It aims for asymmetry: make cheap automation economically pointless.

Strong against

• Frame scraping and one-shot OCR
• Batch classifiers that treat the web as a pile of PNGs
• Brute force on a repeated static background
• High-volume bots that need low cost per solve

Does not claim to stop

• Human solving farms
• Screen recording plus bespoke temporal ML
• Determined adversaries with custom video models

That trade-off is intentional. A captcha cannot stop everything without stopping everyone. The goal is efficiency: block automation at scale, accept that individual motivated solves may still occur, and use TTL, rate limits, server-side verification, and optional API keys to keep residual risk small.

For accessibility, offer an alternative verification path where motion-based challenges are not suitable.

---

Architecture

Browser / any app          BotStop API (Python)         Storage
     |                              |                            |
     |  POST /v1/challenges         |  create GIF + record       |
     |----------------------------->|---------------------------->|
     |  GET  .../animation.gif      |                            |
     |<-----------------------------|                            |
     |  POST .../verify             |  HMAC + TTL check          |
     |----------------------------->|                            |

• Python package — generator, verifier, API server
• @botstop/client — npm/pnpm client and drop-in widget
• Built-in — refresh, TTL, rate limits, optional API key
• Language-agnostic — any stack that can POST JSON and display a GIF

Install (Python)

pip install botstop[api]

# development
pip install -e ".[api,dev]"

Run the API

# set secrets first (see .env.example)
botstop serve

# or bind publicly
botstop serve --host 0.0.0.0 --port 8787

OpenAPI docs: http://127.0.0.1:8787/docs

Environment

Variable	Default	Purpose
BOTSTOP_SECRET	local-dev-secret	HMAC signing key (change in production)
BOTSTOP_API_KEY	(empty)	Require X-API-Key header when set
BOTSTOP_TTL_SECONDS	300	Challenge expiry
BOTSTOP_CORS_ORIGINS	*	Comma-separated allowed origins
BOTSTOP_RATE_LIMIT	60	Requests per IP per minute
BOTSTOP_STORAGE_DIR	.botstop-data	GIF storage

HTTP API

POST /v1/challenges
→ { "challenge_id", "gif_url", "digit_length", "expires_in" }

GET /v1/challenges/{id}/animation.gif
→ image/gif

POST /v1/challenges/refresh
{ "previous_id": "optional-old-id" }
→ new challenge payload

POST /v1/challenges/{id}/verify
{ "answer": "482913" }
→ { "ok": true, "reason": "ok" }

JavaScript / TypeScript

Build the client package:

cd client
pnpm install
pnpm build

Use in your app:

pnpm add @botstop/client

import { BotStopClient, mountBotStopWidget } from "@botstop/client";

const client = new BotStopClient({
  baseUrl: "http://127.0.0.1:8787",
  apiKey: import.meta.env.VITE_BOTSTOP_API_KEY,
});

mountBotStopWidget(client, {
  target: document.getElementById("captcha")!,
  onVerified: (result) => {
    if (result.ok) submitForm();
  },
});

CLI (local / dev)

botstop generate --reveal-answer
botstop verify --bundle outputs\<id>.bundle.json --answer 482913

Python library

Embed generation and verification in your own backend:

from botstop import create_captcha, verify_captcha

result = create_captcha(demo_html=False)
check = verify_captcha(bundle_path=result.bundle_path, submitted_answer="482913")

Production checklist

1. Set a strong BOTSTOP_SECRET
2. Set BOTSTOP_API_KEY and pass it from your frontend or backend proxy
3. Restrict BOTSTOP_CORS_ORIGINS to your domain
4. Run behind HTTPS (reverse proxy)
5. Use demo_html=False — never ship answers to the browser
6. Proxy API calls through your backend if the API key should not live in the client

Build

python -m pip install build
python -m build

Outputs: dist/botstop-0.3.0-py3-none-any.whl

License

MIT — see LICENSE.