Cryptographically Verifiable Runtime Integrity for Agentic AI Systems.
LICITRA provides tamper-evident runtime audit trails for autonomous AI agents using hash-chained semantic events, PostgreSQL persistence, and forensic evidence exports.
This repository contains the core FastAPI backend and SDK primitives.
Evidence bundles and reproducible experiments live in the companion repository:
👉 https://github.com/narendrakumarnutalapati/licitra-evidence
- Deterministic canonical JSON hashing
- Per-organization hash-chained event ledger (SHA-256)
- PostgreSQL persistence
- Verification endpoint (
/verify/{org_id}) - Evidence bundle export (
/evidence/{org_id}) - PDF governance reports (
/evidence/{org_id}/pdf) - Multi-organization isolation
- Dev-only tamper/reset endpoints gated by
DEV_MODE - Rollback v0 decision primitive
backend/— FastAPI service + database modelssdk/— minimal client SDKtest-vectors/— deterministic hashing test vectorsarchitecture/— diagrams and design notes.env.example— environment template
- Python 3.12+
- PostgreSQL 16
- Windows PowerShell (or Bash)
- Git
git clone https://github.com/narendrakumarnutalapati/licitra-core.git
cd licitra-core
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r backend/requirements.txtSet environment variables:
$env:DATABASE_URL="postgresql+psycopg://user:pass@localhost:5432/licitra"
$env:DEV_MODE="1"Run server:
python -m uvicorn backend.app.main:app --host 127.0.0.1 --port 8000Server runs at:
POST /eventsGET /verify/{org_id}GET /evidence/{org_id}GET /evidence/{org_id}/pdf
POST /dev/reset/{org_id}POST /tamper/{org_id}/{event_id}POST /tamper-prev/{org_id}/{event_id}POST /dev/delete/{org_id}/{event_id}
LICITRA focuses strictly on cryptographic runtime integrity primitives:
- Immutable semantic event capture
- Deterministic verification
- Tamper detection
- Evidence export
It intentionally does NOT include:
- UI dashboards
- Authentication
- Policy enforcement
- Compliance workflows
LICITRA is a technical primitive, not a governance platform.
Reproducible experiments, tampering scenarios, and exported JSON/PDF evidence bundles:
👉 https://github.com/narendrakumarnutalapati/licitra-evidence
MIT
