Nasiko is an open runtime for AI agents, written in Rust, that discovers the coding agents already running on your machines, unifies their cost and session reporting, and routes their model traffic through policies you control. The project has collected more than 9,400 stars on GitHub and was actively developed through 2026.
Development teams typically run several coding harnesses at once โ Claude Code, Codex, Cursor, OpenCode โ each with its own dashboard and billing. Nasiko finds those agents without changing their configuration, records what they do in one schema, and decouples the wire protocol a harness speaks from the provider that actually serves it, so an Anthropic-format request can be served by an OpenAI key. On top of that runtime layer it adds an MCP gateway, orchestration, human-in-the-loop approvals and a dashboard for deploying and watching your own agents.
Nasiko is free and open source (Apache-2.0) and is self-hosted; you bring your own model provider accounts and pay those providers directly.
The platform is configured through environment variables and ships with a web dashboard for deploying agents, routing traffic, managing tools and watching traces. Session reporting and routing are independent opt-ins: a team can start with read-only discovery and cost visibility, then enable routing per harness when it is ready to take control of model choice.
It is for engineering organizations whose developers already run multiple AI coding agents and who need one place to see the spend, enforce policy and choose models โ without ripping out the tools developers like.