lexigram-ai-governance

AI governance for the Lexigram Framework — policy enforcement, audit trails, budget tracking

Overview

AI usage governance for the Lexigram Framework. Enforces budget caps, rate limits, and model access policies on LLM requests — with a full audit trail, soft-limit callbacks, TPM/cost sliding windows, and hot-reloadable configuration. Zero-config usage starts with sensible defaults.

Install

uv add lexigram-ai-governance

Quick Start

from lexigram import Application
from lexigram.di.module import Module, module

from lexigram.ai.governance import GovernanceModule
from lexigram.ai.governance.config import GovernanceConfig

@module(imports=[
    GovernanceModule.configure(
        GovernanceConfig(
            monthly_budget=50.0,
            enforce_budget=True,
            soft_limit_pct=0.8,
            rpm_limit=60,
            restricted_models=["gpt-4o"],
        )
    )
])
class AppModule(Module):
    pass

app = Application(modules=[AppModule])
if __name__ == "__main__":
    app.run()

Configuration

Zero-config usage: Call GovernanceModule.configure() with no arguments to use defaults.

Option 1 — YAML file

# application.yaml
ai_governance:
  enabled: true
  monthly_budget: 100.0
  enforce_budget: true
  soft_limit_pct: 0.8
  rpm_limit: 60

Option 2 — Profiles + Environment Variables (recommended)

export LEX_AI_GOVERNANCE__MONTHLY_BUDGET=100.0
# Environment variables for each field

Option 3 — Python

from lexigram.ai.governance.config import GovernanceConfig
from lexigram.ai.governance import GovernanceModule

config = GovernanceConfig(
    monthly_budget=100.0,
    enforce_budget=True,
    soft_limit_pct=0.8,
    rpm_limit=60,
)
GovernanceModule.configure(config)

Config reference

Module Factory Methods

Key Features

• Budget enforcement: Monthly budget caps with soft-limit callbacks
• Rate limiting: RPM and TPM sliding windows via BudgetTracker
• Model access control: Per-user and per-role allowlist/denylist with glob patterns
• Audit trail: Full governance decision recording via AIAuditStore
• Hot reload: Update limits at runtime without restarting
• Persistence backends: In-memory, Redis, and database backends

Testing

async with Application.boot(modules=[GovernanceModule.stub(
    GovernanceConfig(restricted_models=["gpt-4o"])
)]) as app:
    # your test code
    ...

Key Source Files