hiitl 0.1.0

The control point for software that can act. Observe, enforce, and audit every action.

hiitl — Python SDK

The control point for AI agents. Python SDK with embedded policy evaluator.

Install

pip install hiitl

Quick start

from hiitl import HIITL

# Zero config — observe everything, block nothing
hiitl = HIITL()

decision = hiitl.evaluate("send_email", parameters={
    "to": "[e-mail]",
    "subject": "Order update",
})

if decision.allowed:
    send_email(...)

With policy enforcement

hiitl = HIITL(
    agent_id="payment-agent",
    policy_path="./policy.yaml",
    mode="RESPECT_POLICY",
)

decision = hiitl.evaluate("process_payment", parameters={
    "amount": 5000.00,
    "currency": "USD",
})

if decision.allowed:
    process_payment(...)
elif decision.needs_approval:
    queue_for_review(decision)
elif decision.blocked:
    log_blocked(decision)

Hosted mode

hiitl = HIITL(
    agent_id="payment-agent",
    org_id="org_yourcompany1234567",
    api_key="sk_live_...",
    server_url="https://ecp.hiitl.com",
    environment="prod",
    mode="RESPECT_POLICY",
)

# Same evaluate() call, same Decision object
decision = hiitl.evaluate("process_payment", parameters={"amount": 500})

API

HIITL(**kwargs)

All parameters are keyword-only with defaults. HIITL() works with no arguments.

Parameter	Default	Description
agent_id	"default"	Agent identifier
environment	"dev"	dev, stage, or prod
mode	"OBSERVE_ALL"	OBSERVE_ALL or RESPECT_POLICY
policy_path	None	Path to policy file (YAML or JSON)
audit_db_path	"./hiitl_audit.db"	SQLite audit log path
api_key	None	API key for hosted mode
server_url	None	Server URL for hosted mode

hiitl.evaluate(action, **kwargs) -> Decision

Parameter	Required	Description
action	Yes	Action name (e.g., "send_email")
parameters	No	Action parameters dict
target	No	Target resource dict
operation	No	Operation type (default: "execute")
user_id	No	User identifier
sensitivity	No	Sensitivity labels

Decision

Property	Type	Description
.allowed	bool	Can the action proceed?
.decision	str	Decision type (ALLOW, BLOCK, etc.)
.reason_codes	list[str]	Why this decision was made
.policy_version	str	Policy version used
.ok	bool	Alias for .allowed
.blocked	bool	True if BLOCK
.needs_approval	bool	True if REQUIRE_APPROVAL
.observed	bool	True if OBSERVE mode
.would_be	str	What enforce mode would do (OBSERVE only)

Requirements

• Python 3.12+

Documentation

• Python Quickstart
• Full Documentation
• Examples

License

MIT