Skip to content

vorionsys/cognigate

Cognigate

Open enforcement runtime for the BASIS governance standard

INTENT → ENFORCE → PROOF · Live at cognigate.dev

Python FastAPI CI License Stars Discord

This is v0.1 — early, experimental, and explicitly not production-hardened. We built it because we needed it ourselves and wanted to share it humbly with the community.

The Cognigate Engine implements the BASIS (Behavioral Agent Standard for Integrity and Safety) specification through three core governance layers:

Layer Purpose Endpoint
INTENT Goal normalization & risk assessment POST /v1/intent
ENFORCE Policy validation & trust gating POST /v1/enforce
PROOF Immutable cryptographic audit ledger POST /v1/proof

Why Cognigate?

LangChain, CrewAI, and AutoGen give you powerful agent orchestration. Cognigate adds the missing open governance layer:

  • BASIS compliance enforcement — deterministic pre-reasoning checks
  • Real-time ATSF trust tiers (T0–T7) — capability gating by trust score
  • Cryptographic PROOF generation — SHA-256 linked audit chain
  • Seamless wrapping around any agent framework

We know this is early. Help us make it better.

Quick Start

git clone https://github.com/vorionsys/cognigate.git
cd cognigate
pip install -e .
uvicorn app.main:app --reload
# API docs at http://localhost:8000/docs

Live API

Documentation

Usage Example

import httpx

async def check_intent(goal: str, entity_id: str):
    async with httpx.AsyncClient() as client:
        # 1. Normalize intent
        intent = await client.post("https://cognigate.dev/v1/intent", json={
            "entity_id": entity_id,
            "goal": goal,
        })

        # 2. Enforce policies
        verdict = await client.post("https://cognigate.dev/v1/enforce", json={
            "plan": intent.json()["plan"],
            "entity_id": entity_id,
            "trust_level": intent.json()["trust_level"],
            "trust_score": intent.json()["trust_score"],
        })

        return verdict.json()

# Returns: {"allowed": true, "action": "allow", ...}

Trust Tiers (T0–T7)

Tier definitions are sourced from @vorionsys/shared-constants — canonical definitions for the Vorion ecosystem.

Tier Name Score Range Capabilities
T0 Sandbox 0–199 Read-only, no external calls; all intents require approval
T1 Observed 200–349 Basic tools, scoped data; enhanced logging active
T2 Provisional 350–499 Standard tools, rate-limited; sensitive ops require review
T3 Monitored 500–649 Full standard toolset; continuous monitoring
T4 Standard 650–799 Extended tools + external APIs; green-light for most operations
T5 Trusted 800–875 Cross-namespace access; elevated authority scope
T6 Certified 876–950 Administrative operations; can approve others' intents
T7 Autonomous 951–1000 Unrestricted within policy; self-governing

Trust scores use asymmetric dynamics: failures apply a tier-scaled 7–10× penalty while positive signals provide smaller recovery bonuses. Idle agents decay to 50% at 182 days (182-day half-life). See BASIS spec for full scoring model.

Test Coverage

Cognigate's Python test suite covers compliance verification across 13 frameworks:

Framework Tests Coverage
NIST SP 800-53 20 52 controls implemented
SOC 2 13 Type II controls
PCI DSS 32 Controls mapped
FedRAMP 35 Controls mapped
ISO 27001, GDPR, EU AI Act, + 6 more ~15 Framework health checks

Total Python tests: 692 | Run: pytest tests/

Vorion Ecosystem

Component Description
Vorion monorepo Full governance stack (BASIS, ATSF, CAR, SDKs)
@vorionsys/atsf-core TypeScript trust scoring SDK
vorion.org Main site
cognigate.dev Live runtime demo

Governance

The people behind Cognigate

Created by Alex Blanc and Ryan Cason (Bo Xandar Lee) — two former banquet servers who taught themselves to code with AI and wanted to give something back to the community.

Read our full story →

Support

License

Apache-2.0 — see LICENSE for details.


Built with BASIS · vorion.org · Feedback welcome — even the brutal kind.