sovigl 0.1.10

SOVIGL returns approved, pending, or blocked for any action.

SOVIGL

Policy enforcement and audit layer for AI agents.

Every action your AI takes - governed before it executes. pip install sovigl

What it does

SOVIGL sits between your AI agent and the action it wants to take. Before anything executes - payment, approval, data operation

• SOVIGL evaluates it against your policy and returns one of three outcomes:

• APPROVED - executes immediately

• PENDING - held for human approval

• BLOCKED - stopped permanently

Every decision is permanently recorded with a unique ID, plain English explanation, risk score, and the exact policy version that triggered it.

Demo credentials & base URL

• org_id: demo_fintech by default (the server can override via SOVIGL_DEMO_ORG_ID).
• api_key: demo_key_123 is no longer a supported demo default. Use the same plaintext as Railway variable SOVIGL_DEMO_API_KEY so it matches the hash stored for the demo org. Recommended stable value: sk_sovigl_demo_2026_stable (or whatever you set in Railway—paste that value into sovigl.configure() and examples).
• Base URL: https://web-production-e334b.up.railway.app is the current public deployment used in examples. Replace with your own SOVIGL_BASE_URL / custom domain when not using this demo. Railway URLs can change if you recreate the service—pass base_url to configure() if needed.

Quick start

import sovigl

sovigl.configure(api_key="your-key", org_id="your-org")

decision = sovigl.evaluate(
    action="payment.create",
    context={
        "amount": 5000,
        "role": "employee",
        "user_id": "user_123",
        "agent_id": "invoice_bot"
    }
)

print(decision.status)               # approved
print(decision.decision_id)          # permanent audit ID
print(decision.reason)               # why it was approved
print(decision.risk_assessment)      # risk score 0.0-1.0
print(decision.explanation_registry) # full explainability
print(decision.policy_version)       # which policy version
print(decision.approved)             # True

Demo - no signup needed

Try it instantly with the demo org. Defaults use demo_fintech and the recommended Railway demo key sk_sovigl_demo_2026_stable when that value is set in SOVIGL_DEMO_API_KEY:

import sovigl

# Optional: explicit configure if your Railway demo key differs
# sovigl.configure(api_key="sk_sovigl_demo_2026_stable", org_id="demo_fintech")

decision = sovigl.evaluate(
    action="payment.create",
    context={"amount": 5000}
)

print(decision.status)      # approved
print(decision.decision_id) # permanent ID
print(decision.reason)      # plain English reason

Live dashboard: https://web-production-e334b.up.railway.app/dashboard

decision_id, /check, and /evaluate

The Python SDK calls POST /check. The backend expects a decision_id on each request (idempotency key). The SDK generates a new UUID for every call unless you pass decision_id in context. Use a new id per new evaluation; reuse only when you intend a retry or idempotent replay of the same logical decision.

On the server, POST /evaluate uses the same action model and also requires decision_id in the body. For org_id == demo_fintech, /evaluate may allow no api-key (demo-only). Non-demo orgs still require a valid API key. The SDK still sends api-key by default for /check, which matches normal production use.

Full response object

evaluate() returns a Decision object:

Field	Type	Description
status	string	approved / pending / blocked
decision_id	string	permanent immutable audit ID
reason	string	why this decision was made
explanation_registry	dict	full explainability record
risk_assessment	dict	risk score 0.0-1.0 + factors
policy_version	int	which policy version was active
approval_id	string	human approval reference
summary	string	plain English summary
approved	bool	True if status == approved
pending	bool	True if status == pending
blocked	bool	True if status == blocked
raw	dict	complete backend response

What makes it different

• Pre-execution gate - nothing executes without a decision. Not post-execution logging.

• Policy versioning - auditors know exactly which rule was active at decision time.

• Business policy engine - amount thresholds, role-based routing, mandate enforcement, fraud detection built in.

• Compliance evidence - every decision automatically satisfies EU AI Act Art.12/13/14, MAS FEAT, NIST AI RMF, RBI FREE-AI.

• Fully hosted - no self-hosting needed. Works in an afternoon.

Context fields

Field	Type	Description
amount	number	Transaction amount
role	string	employee / manager / admin
user_id	string	Who is performing the action
agent_id	string	AI agent identifier
description	string	Optional note
decision_id	string	Optional; idempotency key (new UUID per call if omitted)

Action types

Any string in dot notation:

• payment.create
• expense.submit
• transfer.initiate
• loan.approve
• vendor.onboard
• data.export

Pending workflow

decision = sovigl.evaluate(
    action="payment.create",
    context={"amount": 25000, "role": "employee"}
)

if decision.pending:
    # Route to human approver
    print(f"Approval required: {decision.approval_id}")
    print(f"Reason: {decision.reason}")

Node.js

const sovigl = require("./node/index.js");

const decision = await sovigl.evaluate({
    action: "payment.create",
    context: { amount: 5000, role: "employee" }
});

console.log(decision.status);      // approved
console.log(decision.decision_id); // permanent audit ID
console.log(decision.reason);      // why

Production access

Email [e-mail] to get your org_id and api_key.

Free during beta. No credit card. No commitment.

Links

• Dashboard: https://web-production-e334b.up.railway.app/dashboard
• PyPI: https://pypi.org/project/sovigl/
• Backend repo: https://github.com/riteshkumar10000/SOVIGL
• This SDK repo: https://github.com/riteshkumar10000/sovigl-sdk