drako 2.5.3

The Python Governance Platform for AI Agents — compliance middleware for LangGraph, CrewAI, AutoGen

Drako SDK

The Trust Layer for AI Agents — Scan for governance gaps offline. Enforce policies in production via the Drako platform.

---

Quick Start

1. Install

pip install drako

With framework-specific extras:

pip install drako[crewai]      # CrewAI support
pip install drako[langgraph]   # LangGraph support
pip install drako[autogen]     # AutoGen support
pip install drako[all]         # All frameworks

2. Scan Your Project (free, offline, no account needed)

drako scan .
# → Governance Score: 35/100 | 8 findings | Agent BOM: 3 agents, 12 tools
# → Run `drako auth login` to enable runtime governance

3. Connect to Platform (optional, enables runtime governance)

drako init

This will:

• Validate your API key
• Generate .drako.yaml config
• Create a framework-specific middleware file
• Enable runtime policy enforcement

4. Integrate Runtime Middleware

CrewAI

from drako import with_compliance
from my_crew import MyCrew

crew = with_compliance(MyCrew())
result = crew.kickoff()

LangGraph

from drako import with_langgraph_compliance

graph = with_langgraph_compliance(my_graph.compile())
result = graph.invoke(input)

AutoGen

from drako import with_autogen_compliance

chat = with_autogen_compliance(my_group_chat)

Direct Client

from drako import DrakoClient

client = DrakoClient.from_env()
result = client.verify_agent_identity_sync("my-agent", "researcher")

---

CLI Commands

Command	Description	Requires Account
drako scan .	Scan project for governance gaps	No
drako init	Initialize runtime SDK in your project	Yes
drako status	Check connection and quota status	Yes
drako verify	Verify audit hash chain integrity	Yes
drako serve	Start a local MCP compliance server	Yes

---

Configuration

.drako.yaml:

version: "1.0"
tenant_id: your_tenant_id
api_key_env: DRAKO_API_KEY
endpoint: https://api.getdrako.com
framework: crewai

tools:
  audit_log_action: true
  verify_agent_identity: true
  evaluate_policy: true

guardrails:
  prompt_injection_detection: true
  dlp_scanning: false

trust:
  enabled: true
  decay_half_life_hours: 168
  circuit_breaker_threshold: 3

---

Capabilities

Scan CLI (free, offline)

Capability	Status	Details
Governance Score	✅ Built-in	25 rules, 4 categories, deterministic evaluation
Agent BOM	✅ Built-in	AST-based discovery, 6 frameworks
EU AI Act Gap Detection	✅ Built-in	Art. 9, 11, 12, 14 mapping with fix snippets
SARIF 2.1.0	✅ Built-in	GitHub Code Scanning compatible
SVG Badge	✅ Built-in	Embeddable governance badge

Runtime Platform (requires account)

Capability	Status	Details
Policy Enforcement	✅ Production	Real-time evaluation on tool calls
DLP/PII Detection	✅ Production	Presidio-based, 8+ entity types, anonymization
Circuit Breaker	✅ Production	State machine + EigenTrust scoring
Audit Trail	✅ Production	SHA-256/BLAKE3 hash chain + digital signatures
Trust Score	✅ Production	0-100 dynamic score per agent
OTEL Export	🔜 Next Sprint	OpenTelemetry semantic conventions

---

Environment Variables

Variable	Description	Required
DRAKO_API_KEY	Your API key (for runtime features)	For platform
DRAKO_ENDPOINT	Custom backend URL	No
DRAKO_TENANT_ID	Override tenant ID	No

---

Development

git clone https://github.com/angelnicolasc/drako.git
cd drako
pip install -e ".[dev]"
pytest

---

License

MIT