captchakit

Production-ready, async-first captcha library for Python 3.10+. Zero runtime deps beyond Pillow. Drop-in adapters for FastAPI, aiogram, Discord and Django — plus Redis / Postgres storage, rate limiting and Prometheus metrics.

---

Showcase

Four built-in themes, plus SVG and audio renderers — all produced by the same CaptchaManager API.

Theme.CLASSIC	Theme.DARK	Theme.PASTEL	Theme.HIGH_CONTRAST · WCAG AA
MathChallengeFactory → ImageRenderer		WordChallengeFactory → SVGRenderer (~2 KB)

Accessibility: accessibility.wav — an AudioRenderer sample (~110 KB WAV) that dictates the same solution as a visual challenge, so screen-reader users get an a11y-friendly alternative.

All assets above were rendered by scripts/render_showcase.py using the public API — run it yourself to regenerate them.

Why captchakit

Feature	lepture/captcha	claptcha	multicolorcaptcha	captchakit
Async API	❌	❌	❌	✅
py.typed + mypy strict	⚠️	❌	❌	✅
TTL & attempt tracking	❌	❌	❌	✅ built-in
Pluggable storage	❌	❌	❌	✅ Protocol — Memory / Redis / Postgres
Pluggable rate limiter	❌	❌	❌	✅ Protocol — in-memory & Redis token-bucket
Prometheus metrics	❌	❌	❌	✅ opt-in
Framework adapters	❌	❌	❌	✅ FastAPI · aiogram · Discord · Django
Audio challenge (a11y)	❌	❌	❌	✅ AudioRenderer
i18n prompt hooks	❌	❌	❌	✅ en / tr / de / es + custom catalog
Core runtime deps	+Pillow	+Pillow	+Pillow	+Pillow

Install

pip install captchakit                  # core

# adapters
pip install "captchakit[fastapi]"
pip install "captchakit[aiogram]"
pip install "captchakit[discord]"
pip install "captchakit[django]"

# storage
pip install "captchakit[redis]"         # + rate-limit token bucket
pip install "captchakit[postgres]"

# observability
pip install "captchakit[metrics]"       # Prometheus adapter

30-second example

import asyncio
from captchakit import (
    CaptchaManager, ImageRenderer, MemoryStorage, TextChallengeFactory,
)

async def main() -> None:
    manager = CaptchaManager(
        factory=TextChallengeFactory(length=5),
        renderer=ImageRenderer(),
        storage=MemoryStorage(),
        ttl=120.0,
        max_attempts=3,
    )
    challenge_id, png_bytes = await manager.issue()
    # ... show png_bytes to the user, receive their answer ...
    ok = await manager.verify(challenge_id, user_input="ABCDE")
    print("verified" if ok else "wrong answer, more attempts remain")

asyncio.run(main())

FastAPI in 10 lines

from fastapi import Depends, FastAPI
from captchakit import CaptchaManager, ImageRenderer, MathChallengeFactory, MemoryStorage
from captchakit.adapters.fastapi import captcha_router, verify_captcha

manager = CaptchaManager(MathChallengeFactory(), ImageRenderer(), MemoryStorage())
app = FastAPI()
app.include_router(captcha_router(manager, prefix="/captcha"))

@app.post("/register")
async def register(_: None = Depends(verify_captcha(manager))) -> dict[str, bool]:
    return {"ok": True}

Run the bundled demo:

uv run python -m uvicorn examples.fastapi_login:app --reload
# open http://127.0.0.1:8000

Architecture

┌──────────────────────────────────────────────────────────────┐
│  Adapters (FastAPI / aiogram / Discord / Django)             │
├──────────────────────────────────────────────────────────────┤
│  CaptchaManager  (issue → render → persist · verify · TTL)   │
├──────────────┬──────────────┬───────────────┬────────────────┤
│  Challenge   │  Renderer    │  Storage      │  Rate limiter  │
│  Text · Math │  Image · SVG │  Memory       │  NoOp          │
│  Grid · Word │  Audio       │  Redis · PG   │  Token bucket  │
└──────────────┴──────────────┴───────────────┴────────────────┘
             ↓ i18n translator · metrics sink · clock

Everything coloured here is a Protocol — drop in your own implementation without subclassing.

• Constant-time comparison via hmac.compare_digest.
• Crypto-safe randomness via secrets for solution generation.
• CPU-bound Pillow drawing offloaded to a worker thread with asyncio.to_thread.
• Multi-process safe when paired with RedisStorage or PostgresStorage.

Performance

Operation	Mean (ms)	Median (ms)	p99 (ms)
ImageRenderer.render (PNG)	3.64	3.54	4.84
SVGRenderer.render (SVG)	0.03	0.03	0.05
AudioRenderer.render (WAV)	2.69	2.60	3.72
issue + verify round-trip	2.73	2.57	5.29

Measured on a single CPython 3.13 thread, Windows 10, 500 iterations after 20 warmups. Reproduce with uv run python benchmarks/bench.py.

Production deployment

• Run behind a proper reverse proxy with TLS and WAF rules.
• Use RedisStorage or PostgresStorage if you scale beyond a single worker — MemoryStorage is per-process.
• Wire RateLimiter to RedisTokenBucket (or your edge WAF) when exposing the issue endpoint publicly.
• Expose PrometheusMetrics on :9090/metrics and alert on captchakit_too_many_attempts_total spikes.
• Pair at least one visual renderer with AudioRenderer for accessibility.
• Set ttl short (60–180 s) and max_attempts low (2–3) — captchakit is a raise-the-cost layer, not a fortress.

Security scope

captchakit is a lightweight human-check — it raises the cost for casual spam and scripted abuse. It is not a bot-farm-grade defence.

For high-value forms (payment, password reset, account takeover) use hCaptcha, Cloudflare Turnstile or reCAPTCHA Enterprise in addition to captchakit, and enforce rate limiting at the edge of your application.

Vulnerability reports: see SECURITY.md.

Stability & compatibility

• 1.x is API-stable — public symbols follow semver. See docs/stability.md for the policy.
• Supported Python versions: 3.10 → 3.13. New 3.x is added within one MINOR of upstream release.
• Every MINOR gets security patches throughout the life of the 1.x line.

Documentation

Full docs at https://akerem16.github.io/captchakit/ — quickstart, adapter guides, storage & rate-limit recipes, i18n, metrics, accessibility, API reference.

For AI agents / LLM tooling

captchakit ships first-class context files so coding agents (Cursor, Windsurf, Claude Code, Copilot Workspace, aider, …) can integrate it correctly on the first try:

• llms.txt — concise, llmstxt.org-compatible index of every public symbol and the canonical recipes.
• AGENTS.md — deterministic, copy-paste task templates for common asks (“add a captcha to my FastAPI route”, “swap storage to Redis”, etc.), plus the API contract and pitfalls to avoid.

Drop either file into your agent's retrieval / context window and it should be able to wire captchakit into an app without hallucinating imports or inventing kwargs. Minimal hello-world an agent can output verbatim:

# Install:  pip install "captchakit[fastapi]"
from fastapi import Depends, FastAPI
from captchakit import CaptchaManager, ImageRenderer, MathChallengeFactory, MemoryStorage
from captchakit.adapters.fastapi import captcha_router, verify_captcha

manager = CaptchaManager(MathChallengeFactory(), ImageRenderer(), MemoryStorage())
app = FastAPI()
app.include_router(captcha_router(manager, prefix="/captcha"))

@app.post("/signup")
async def signup(_: None = Depends(verify_captcha(manager))) -> dict[str, bool]:
    return {"ok": True}

Contributing

Contributions welcome. Local development:

git clone https://github.com/akerem16/captchakit
cd captchakit
uv sync --all-extras
uv run ruff check .
uv run mypy
uv run pytest
uv run bandit -c pyproject.toml -r src
uv run pip-audit

See CONTRIBUTING.md.

License

MIT — see LICENSE.