AuthGraph 0.3.1

Execution intelligence for autonomous systems — advisory, not authorization truth.

AuthGraph discovers execution surfaces, maps architecture, scores risk, and recommends policy-preserving remediation. TrigGuard remains the sole authorization authority (PERMIT / DENY / ESCALATE / SILENCE).

pip install authgraph==0.3.1

architecture → analyze → install → protect → simulate → explain → score → recommend
                              ↓
                        human review
                              ↓
                     TrigGuard authorize

0.3.1 does not rewrite repositories. No patch / apply / undo.

from authgraph import Plan, simulate, analyze_repository, guard

# Discovery (read-only)
report = analyze_repository(".")

# Simulation asks TrigGuard — never executes callables
plan = Plan()
plan.add("deploy.release", ship, title="ship")
print(simulate(plan))

CLI (advisory):

authgraph architecture .
authgraph analyze .
authgraph install .          # writes .authgraph/ plan only
authgraph protect .          # human review of plan (no source edits)
authgraph score .
authgraph recommend .        # structured remediation — never bypass suggestions
authgraph report .
authgraph simulate examples/deployment.py
authgraph explain <receipt> --action deploy.release
authgraph doctor .
authgraph why AG-0001

Hard guarantees:

Requires TRIGGUARD_API_KEY, TRIGGUARD_ORG_ID for live simulation/authorization. Gateway default: https://api.trigguardai.com.

Full guide: docs/QUICKSTART.md · Policy preservation: docs/POLICY_PRESERVATION.md

APIs

from authgraph import guard, authorization, run_step, run_step_async, run_many
from authgraph import Plan, simulate, explain, analyze_repository

Not a CrewAI/LangGraph/Quanta product. Optional integration hooks live under authgraph.integrations and do not bundle external runtimes.

Demos: examples/demos/

License

Apache-2.0 · TrigGuard AI