aigc-sdk

AIGC — Auditable Intelligence Governance Contract

AIGC is a Python SDK for deterministic, fail-closed governance of AI model invocations. It validates every invocation against a declared policy, enforces role and schema constraints, evaluates optional custom gates and risk scoring, and emits a tamper-evident audit artifact for every pass or fail path.

Governance in AIGC is runtime enforcement, not documentation and not prompting.

At a Glance

• Package: pip install aigc-sdk
• Import: import aigc
• Current release: v0.3.3 on 2026-04-10
• Current release scope: workflow-aware governance, audit schema v1.4, AuditLineage, ProvenanceGate, RiskHistory, @governed defaults to split enforcement
• Verification baseline: 966 tests, coverage above the 90% CI gate

Why This Repo Exists

Most AI governance guidance stays advisory. AIGC is opinionated about turning governance into an executable contract:

• Policies are declarative YAML and validated against JSON Schema.
• Enforcement is deterministic and fail-closed.
• Audit artifacts are mandatory, structured, and checksum-based.
• Governance is model- and provider-agnostic.
• Split enforcement is available when hosts need pre-call authorization before token spend.

If a model call cannot be justified by policy and evidenced by an audit record, AIGC treats it as invalid.

Runtime Model

AIGC sits at the invocation boundary between the host application and the model provider.

1. The host assembles an invocation and selects a policy.
2. AIGC loads and resolves the policy, enforces ordered governance gates, and computes audit metadata.
3. AIGC returns or emits a PASS/FAIL audit artifact that can be stored, exported, chained, or inspected offline.

Since v0.3.3, split enforcement is the default — Phase A runs before the model call, Phase B validates output after. Pass pre_call_enforcement=False for the legacy unified mode (deprecated).

Release Narrative

This is the versioned story of the repo's current state and how it evolved release by release.

For the full change log, use CHANGELOG.md.

Installation

pip install aigc-sdk

Editable install from source:

python3 -m venv aigc-env
source aigc-env/bin/activate
python -m pip install --upgrade pip setuptools wheel
pip install --no-build-isolation -e '.[dev]'

The --no-build-isolation flag keeps editable installs working in network-restricted environments by reusing already-installed build tools.

Quick Start

Unified enforcement remains the simplest integration path:

from aigc import enforce_invocation

artifact = enforce_invocation(
    {
        "policy_file": "policies/base_policy.yaml",
        "model_provider": "anthropic",
        "model_identifier": "claude-sonnet-4-6",
        "role": "assistant",
        "input": {"query": "Summarize this incident"},
        "output": {"result": "Summary text", "confidence": 0.94},
        "context": {"role_declared": True, "schema_exists": True},
    }
)

For new code that needs isolated configuration, use the instance API:

from aigc import AIGC, JsonFileAuditSink

engine = AIGC(sink=JsonFileAuditSink("audit.jsonl"))
artifact = engine.enforce(invocation)

Split Enforcement in v0.3.2

Split mode lets you authorize before the model call and validate output after the call:

from aigc import enforce_post_call, enforce_pre_call

pre = enforce_pre_call(
    {
        "policy_file": "policies/base_policy.yaml",
        "model_provider": "anthropic",
        "model_identifier": "claude-sonnet-4-6",
        "role": "assistant",
        "input": {"query": "Summarize this incident"},
        "context": {"role_declared": True, "schema_exists": True},
    }
)

output = model.generate(...)
artifact = enforce_post_call(pre, output)

The @governed decorator uses split enforcement by default (since v0.3.3):

from aigc import governed

@governed(
    policy_file="policies/base_policy.yaml",
    role="assistant",
    model_provider="anthropic",
    model_identifier="claude-sonnet-4-6",
)
def run_model(input_data, context):
    return model.generate(input_data)

Phase A runs before the model call; Phase B validates output after. Pass pre_call_enforcement=False for legacy unified mode (deprecated).

Interactive Demo

The maintained demo is a React frontend plus FastAPI backend:

• Site: https://nealsolves.github.io/aigc/
• Coverage: 7 labs across risk scoring, signing, audit chain, composition, loaders and policy dates, custom gates, and compliance export
• Purpose: hands-on orientation to the runtime that shipped in v0.3.x

CLI Surface

The aigc console script exposes three practical commands:

• aigc policy lint <file...> for syntax and schema checks
• aigc policy validate <file...> for semantic validation, including composition and cycle detection
• aigc compliance export --input audit.jsonl [--output report.json] [--lineage] for offline compliance reporting over stored audit trails; add --lineage to include DAG-level lineage analysis (node counts, duplicate detection, root/leaf/orphan lists, cycle detection)

Repo Guide

If you are new to the repo, start here:

Development Gates

Before release, the repo expects these checks to pass locally:

python -m pytest --cov=aigc --cov-report=term-missing --cov-fail-under=90
flake8 aigc
npx markdownlint-cli2 "**/*.md"