We Built a Governance CLI for Multi-Agent AI (And Published It to npm)

#ai #mcp #opensource #cli

Most AI CLIs help you code faster. Ours helps you govern 13 agents that run a business.

The Problem

We run 13 AI agents on a single VPS. Tamara dispatches agents. Lou scans for grants. Roman publishes content. Kris finds government contracts. Each one can edit files, call APIs, and make decisions.

Without governance, this is chaos. With system prompts alone, it is polite chaos.

What palyan-cli Does

Three layers in one command:

Layer 1 - Governance (20 commands)
Wraps our Nervous System MCP server. Preflight checks before any file edit. Hash-chained audit trails. Drift detection. Kill switch. Security scans.

npx palyan-cli status
npx palyan-cli audit --scope full
npx palyan-cli kill <agent-name>

Layer 2 - Dispatch (10 commands)
Named agent dispatch with budget caps. Deadline tracking. Partner management. Opportunity scanning.

npx palyan-cli agents
npx palyan-cli dispatch roman "Write a blog post about MCP governance"
npx palyan-cli deadlines

Layer 3 - A2A Agent Cards
Every agent has a machine-readable identity card following Google's Agent-to-Agent protocol (v0.3.0). Name, role, capabilities, constraints, permissions.

npx palyan-cli agent-card tamara

Why This Exists

Every MCP governance tool launched in 2026 governs tool calls. They sit between the client and the MCP server like a firewall.

We govern agent behavior. Preflight runs before the edit happens. Drift detection catches when an agent wanders. The audit chain proves what happened. The kill switch stops it.

They build TSA. We build the flight control tower.