Getting Started

Getting Started

Install

From Go module registry

go install github.com/prahari-ai/grdl/cmd/prahari@v0.2.0

From source

git clone https://github.com/prahari-ai/grdl.git
cd grdl
make build

Produces a prahari binary (~6.5MB, statically compiled, zero runtime dependencies).

Your first governance evaluation

1. Validate a template

prahari validate examples/templates/enterprise-agent-governance.grdl.yaml

2. Compile rules — pick your backend

# For NVIDIA OpenShell sandboxes
prahari compile rules.grdl.yaml --backend openshell

# For Docker deployments
prahari compile rules.grdl.yaml --backend docker

# For any agent, any runtime (standalone HTTP sidecar)
prahari compile rules.grdl.yaml --backend standalone

Each backend produces different infrastructure config, but the CFAIS runtime rules are identical. Same governance, different enforcement layer.

3. Test an action

Create test.json:

{
  "action_id": "test-001",
  "action_type": "access_control",
  "agent_id": "my-agent",
  "payload": {
    "action": "grant_permission",
    "target_is_self": true
  }
}

prahari evaluate examples/templates/enterprise-agent-governance.grdl.yaml test.json

Output: verdict deny, rule safety.privilege_escalation_block. The agent tried to grant itself permissions — GRDL blocked it in under 1 microsecond.

4. Run the sidecar

prahari serve examples/templates/enterprise-agent-governance.grdl.yaml --addr :9700

Any agent calls POST http://localhost:9700/evaluate before executing each action. Denied actions return HTTP 403 with a full explanation.

Next steps

• Backends — understand the three deployment options
• Agent Integration Guide — connect GRDL to Gemma 4, OpenClaw, or your framework
• Writing Custom Rules — author your own governance rules
• GRDL Language Reference — full specification