quantumcatena-sentinel 0.1.0

Real-time security layer for AI agents — intercept, evaluate, audit with post-quantum proof

QuantumCatena Sentinel SDK

Real-time security layer for AI agents.
Intercept, evaluate, and audit every action before execution —
with post-quantum cryptographic proof (Dilithium-3 NIST FIPS 204).

---

Install

pip install quantumcatena-sentinel

Quickstart — 3 lines

from sentinel import SentinelClient, BlockedBySentinel

sentinel = SentinelClient(api_key="sk-...", agent_id="my-agent")

@sentinel.protect(action_type="send_data", data_class="pii")
def send_email(recipient: str, subject: str, body: str) -> str:
    return smtp.send(recipient, subject, body)   # only runs if Sentinel allows it

try:
    send_email("[e-mail]", "Client export", "PII data...")
except BlockedBySentinel as e:
    print(f"Blocked. Risk: {e.risk_score}. Reason: {e.reasoning}")

That is all. Your agent cannot bypass Sentinel.

---

How it works

Your Agent
    |
    v
sentinel.protect()  <-- SDK intercepts here, BEFORE execution
    |
    v
Sentinel API  -->  Policy Engine  -->  Risk Scoring  -->  PQC Proof
    |
    v
ALLOW / REVIEW / BLOCK
    |
    v
Your function executes (or BlockedBySentinel is raised)

---

Usage

Decorator pattern

sentinel = SentinelClient(api_key="...", agent_id="invoice-agent")

@sentinel.protect(action_type="read_data", data_class="general")
def read_file(path: str) -> str:
    with open(path) as f:
        return f.read()

@sentinel.protect(action_type="transfer_funds")
def wire_transfer(amount: float, iban: str) -> str:
    return bank.transfer(amount, iban)

Low-level check (inspect before acting)

decision = sentinel.check(
    action_type="transfer_funds",
    target="https://api.stripe.com",
    payload={"amount": 50000, "iban": "DE89..."},
)
print(decision["decision"])     # allow | review | block
print(decision["risk_score"])   # 0.0 - 1.0
print(decision["pqc_proof"])    # Dilithium-3 proof (Enterprise)

SentinelToolkit — for multi-tool agents

from sentinel import SentinelClient, SentinelToolkit

sentinel = SentinelClient(api_key="...", agent_id="my-agent")
toolkit  = SentinelToolkit(sentinel)

@toolkit.tool(action_type="send_data", data_class="internal")
def send_email(recipient: str, subject: str) -> str:
    """Send an internal email."""
    return smtp.send(recipient, subject)

# Pass to Anthropic agent
response = client.messages.create(
    tools=toolkit.anthropic_tools,   # auto-generated schemas
    ...
)

LangChain integration

from sentinel.middleware import SentinelToolkit

toolkit  = SentinelToolkit(sentinel)
lc_tools = toolkit.wrap_langchain([search, email, file_reader])

agent = initialize_agent(lc_tools, llm, ...)

---

Action types

action_type	Typical use
read_data	File read, DB query
send_data	Email, SMS, Slack, webhook
call_api	External HTTP call
execute_code	Shell, Python exec, subprocess
transfer_funds	Payment, bank wire, crypto transfer
write_data	File write, DB insert/update
delete_data	File delete, DB delete

Data classes

data_class	Examples
general	Public info, logs, system metadata
internal	Internal docs, employee emails
pii	Names, emails, IDs, medical data
confidential	Trade secrets, keys, financial records

---

Configuration

sentinel = SentinelClient(
    api_key   = "sk-...",         # or SENTINEL_API_KEY env var
    agent_id  = "prod-agent-1",
    api_url   = "https://sentinel.yourdomain.com",  # self-hosted / Enterprise
    fail_open = True,             # allow if Sentinel unreachable (default)
    verbose   = True,             # print decisions to stdout
)

---

Pricing

Tier	Price	Decisions/month	PQC Layer
Free	0 EUR	10,000	Ed25519 (sim)
Pro	149 EUR/mo	Unlimited	Dilithium-3
Enterprise	Custom	Unlimited	Dilithium-3 + HSM

Get your API key | Docs

---

Open source

The SDK is Apache 2.0.
The Sentinel backend is source-available for self-hosted Enterprise.
The Openclaw Shield dashboard is MIT.

Live demo: https://quantumcatena.github.io/openclaw-shield
API docs: https://sentinel-t26z.onrender.com/docs
Contact: [e-mail]