Nasiko — Developer Control Plane for AI Agent Governance

Introduction

Nasiko is an open-source developer control plane designed for teams that run multiple AI coding agents. It provides centralized policy enforcement, token usage accounting, and an MCP gateway to manage how agents interact with external tools and services.

What Nasiko Does

• Enforces security policies across multiple AI coding agents
• Tracks and limits token usage per agent, user, or project
• Provides an MCP gateway for centralized tool access control
• Monitors agent activity with audit logging
• Manages API key distribution and rotation

Architecture Overview

Nasiko runs as a lightweight proxy between your agents and their LLM/tool providers. The policy engine evaluates each request against configurable rules before forwarding. Token accounting happens at the proxy layer, giving accurate usage data without relying on provider dashboards. The MCP gateway aggregates multiple MCP servers behind a single authenticated endpoint.

Self-Hosting & Configuration

• Install via pip: pip install nasiko
• Initialize a workspace with nasiko init
• Configure policies in the workspace YAML file
• Set up token budgets per team or project
• Deploy behind your existing reverse proxy for production use

Key Features

• Centralized policy enforcement for agent security
• Real-time token usage tracking and budget limits
• MCP gateway with per-tool access control
• Audit logging for compliance and debugging
• Multi-agent support with per-agent configuration

Comparison with Similar Tools

• Portkey AI Gateway — LLM routing focus; Nasiko adds agent governance and MCP gateway
• LiteLLM — Unified LLM proxy; Nasiko provides broader agent lifecycle management
• Helicone — LLM observability; Nasiko combines observability with policy enforcement
• OpenLIT — Telemetry for LLM apps; Nasiko is purpose-built for agent governance

FAQ

Q: Which agents does Nasiko support? A: Any agent that makes LLM API calls or uses MCP servers, including Claude Code, Codex, and custom agents.

Q: Can I set different budgets per project? A: Yes, token budgets are configurable per agent, user, project, or team.

Q: Does it support multiple LLM providers? A: Yes, the proxy supports OpenAI, Anthropic, and any OpenAI-compatible endpoint.

Q: Is it suitable for enterprise use? A: Nasiko is designed for team use with audit logging and RBAC. Evaluate the maturity for your specific compliance requirements.

Sources

• https://github.com/Nasiko-Labs/nasiko