nemesis-shield 0.1.2

Nemesis Shield — Sentinel SDK for Python (web/API/LLM behavioral protection). Learn your app's normal behavior, then block off-baseline requests in enforce mode. ASGI/WSGI/Django middleware + OpenAI/Anthropic/LangChain LLM guards.

nemesis-shield — Python SDK

Native Python SDK for Nemesis Shield. Computes a privacy-preserving request sketch locally, learns your app's normal behavior, and — in enforce mode — blocks off-baseline requests (auth bypass, path traversal, scanners, unusual methods) before your views ever run. Positive-security: this app only ever behaves in these ways. Zero dependencies (stdlib only). Fail-open: if Nemesis is unreachable, your app is unaffected.

pip install nemesis-shield

One line per framework

FastAPI / Starlette (ASGI)

from nemesis_shield.asgi import SentinelMiddleware
app.add_middleware(SentinelMiddleware, token=os.environ["NEMESIS_TOKEN"])

Flask / any WSGI app

from nemesis_shield.wsgi import SentinelWSGI
app.wsgi_app = SentinelWSGI(app.wsgi_app, token=os.environ["NEMESIS_TOKEN"])

Django — add to settings.py and set NEMESIS_TOKEN in the environment:

MIDDLEWARE = ["nemesis_shield.django.SentinelDjango", *MIDDLEWARE]

Raw / anything — use the client directly:

from nemesis_shield import build_sketch
from nemesis_shield.client import SentinelClient
c = SentinelClient(token=os.environ["NEMESIS_TOKEN"])
c.record(build_sketch(method="GET", path="/orders/42", authenticated=True, status=200))

Docs & links

Want more than the one-liner? Full per-framework code, options, and examples:

Framework	Detailed guide & source
FastAPI / Starlette (ASGI)	nemesis_shield/asgi.py
Flask / WSGI	nemesis_shield/wsgi.py
Django	nemesis_shield/django.py
Raw client	nemesis_shield/client.py
LLM guard	nemesis_shield/llm.py
Runnable examples	python/tests

• Full docs & source: https://github.com/eobi/nemesis_shield_sdks/tree/main/python
• Website & console: https://shield.nemesislabs.xyz

How enforcement works

1. Learn — deploy in observe mode (default). Every request is recorded as a privacy-preserving request signature (method + route + params + auth — never bodies or secrets). Review and approve the learned behaviors in the console.
2. Enforce — flip the app to enforce in the console. The SDK polls the compiled allow-list in the background, so no redeploy is needed — the console is the source of truth.
3. Block — any request whose signature isn't in the approved allow-list (once a baseline exists), is explicitly blocked, or matches global threat intel, gets a 403 blocked_by_nemesis_shield before your app handles it. Everything approved passes untouched.

Verified end-to-end (learn → enforce → attack) on FastAPI, Flask, Django and raw: legitimate traffic passes (200) while auth bypass, BOLA, path traversal and scanner probes are blocked (403) and reported.

LLM protection

from nemesis_shield import analyze_llm
result = analyze_llm(prompt=prompt, system=system, response=response, tools=tools)

MIT © Autogon Inc.

Full coverage & safe-unlock

Mount it first / outermost so every route is inspected (not just API routes — attackers hit any path):

app.add_middleware(SentinelMiddleware, token=os.environ["NEMESIS_TOKEN"])  # ASGI; or wrap wsgi_app outermost

What's inspected (privacy-preserving): method + normalized route + query-param structure (names + kinds, never values) + auth flag + status. An off-baseline route, param structure, method, or auth state is blocked in enforce mode. Path-traversal segments normalize to {traversal}.

Safe-unlock (break-glass): the login/auth path is never blocked, so a still-learning baseline can't lock you out. Defaults: /login /signin /sign-in /auth /oauth /session /wp-login.php /wp-admin. Override:

export NEMESIS_SHIELD_BOOTSTRAP="/login,/admin,/healthz"

Verify coverage — in observe mode, hit a normal route, a param, and a scanner path, then confirm all three appear in the console (Activity / Behaviors):

curl -s "http://localhost:8080/" >/dev/null
curl -s "http://localhost:8080/search?q=shoes" >/dev/null
curl -s "http://localhost:8080/.env" >/dev/null   # shows up as an off-baseline behavior