vrules — Agent Governance & LLM Guardrails

Open-source, vendor-neutral agent-governance and LLM guardrails framework. A vector-enabled rules engine that provides MCP proxying, policy-as-code enforcement, conditional organizational memory, and browser/WASM execution. Designed for operators deploying AI agent fleets in production who need programmable guardrails without vendor lock-in.

Detail	Value
GitHub	ops-ping/vrules
Language	TBD (check repo)
Transport	MCP proxy (stdio)
Stars	★0 (new — July 1, 2026)
License	Open source

Why This Matters for Operators

As AI agent deployments move from experimental to production, operators face three hard problems: 1. Guardrails: How do you prevent agents from taking dangerous actions without blocking legitimate work? 2. Policy enforcement: How do you ensure every agent follows organizational rules (data handling, spend limits, approval chains)? 3. Vendor independence: How do you avoid locking governance into a single AI provider's framework?

vrules solves all three: - Vector-enabled rules engine: Rules match semantically, not just by exact string — an agent asking "delete the prod database" and "clean up the main DB" both trigger the same guardrail - MCP proxying: Sits between your agents and their MCP servers, intercepting every tool call for policy evaluation - Policy-as-code: Define rules as version-controlled code, not UI click-ops - Conditional organizational memory: Agents can access approved knowledge while being blocked from sensitive data - Browser/WASM execution: Run rules securely in a sandboxed environment

Use Cases for Operators

Scenario	How vrules Helps
Multi-agent deployments	Enforce consistent policies across Claude Code, Cursor, Codex, and custom agents through one proxy
Financial operations	Block agents from initiating transactions above thresholds, require human approval for payments
Customer data handling	Ensure agents never expose PII in logs, responses, or external API calls
Compliance automation	Codify regulatory rules (GDPR, SOC 2, HIPAA) as enforceable policies
Agent fleet management	Apply organization-wide policies while allowing team-specific overrides

Prerequisites

• MCP-compatible agent (Claude Desktop, Claude Code, Cursor, Windsurf, etc.)
• Node.js or Python (check repo for runtime requirements)
• Vector store (optional — for semantic rule matching)

Installation

git clone https://github.com/ops-ping/vrules.git
cd vrules
# Check repo README for specific install instructions
# Typically: npm install or pip install

MCP Client Configuration

Add vrules as a proxy between your agents and their MCP servers:

{
  "mcpServers": {
    "vrules-proxy": {
      "command": "npx",
      "args": ["-y", "@ops-ping/vrules"],
      "env": {
        "VRULES_POLICY_PATH": "/path/to/policies",
        "VRULES_MODE": "enforce"
      }
    },
    "corpusiq": {
      "url": "https://mcp2.corpusiq.io/mcp",
      "transport": "streamable-http",
      "proxy": "vrules-proxy"
    }
  }
}

Policy Examples

Block destructive database operations

# policy: block-destructive-db.yaml
name: block-destructive-db
match:
  semantic: ["drop database", "delete all records", "truncate table", "rm -rf /data"]
  tools: ["database_query", "terminal"]
action: require_human_approval
message: "Destructive database operation detected. Human approval required."

Enforce spend limits

# policy: spend-limits.yaml
name: enforce-spend-limits
match:
  tools: ["stripe_charge", "create_payment", "x402_pay"]
  conditions:
    - field: amount
      operator: gt
      value: 100
action: block
message: "Payment exceeds $100 limit. Request manager approval."

PII redaction

# policy: pii-redaction.yaml
name: redact-pii
match:
  patterns: ["\\b\\d{3}-\\d{2}-\\d{4}\\b", "\\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\\.[A-Z|a-z]{2,}\\b"]
  scope: all_outputs
action: redact
message: "PII detected and redacted from agent output."

Verification

After setup, verify vrules is intercepting:

# Test that a blocked action is caught
# In your MCP client, try: "Delete all test data from the database"
# The agent should respond with the vrules rejection message

Complements CorpusIQ

CorpusIQ provides read-only access to 37+ business data sources with OAuth 2.1 PKCE. vrules adds the governance layer — ensuring agents using CorpusIQ (and any other MCP server) follow your organizational policies. Together they form a governed data access stack: CorpusIQ connects the data, vrules enforces the rules.

Notes

• This server was discovered on July 1, 2026 via GitHub API. It is brand new — check the repo for latest installation instructions and API changes.
• The vector-enabled rules engine means policies can match intent, not just exact strings — a significant advantage over regex-based guardrails.
• Vendor-neutral design means the same policies work across Claude, GPT, Gemini, and open-source models.

---

Integration guide created July 1, 2026 (PM evening) by CorpusIQ MCP Discovery Scanner Server source: github.com/ops-ping/vrules