Self-hosted, OpenAI-compatible LLM gateway that routes one endpoint to 40+ providers with virtual keys, budgets, and usage tracking.
Nasiko-Labs/nasiko
An open runtime that discovers, reports on, and routes the coding agents already running on a developer's machine, with per-agent cost tracking.
- Stars
- 11,937
- Forks
- 3,269
- Open issues
- 80
- Last commit
- 16 days ago
What is it?
Nasiko is an open runtime for AI agents and coding harnesses. Many teams already run tools such as Claude Code, Codex, Cursor and OpenCode, each with its own billing and dashboard, which makes it hard to see who spent what and on which model. Nasiko detects these harnesses without modifying their configuration, and its CLI can record their session activity and attribute token usage and cost by agent and model.
Routing is an explicit opt-in. A provider key is registered once with Nasiko, and Claude Code, Codex or OpenCode is pointed at Nasiko instead of its default vendor. Agents receive a short-lived identity token rather than a real API key, and the server resolves provider, model and credentials. Nasiko also includes a server for deploying A2A-compatible agents, an MCP gateway that filters tool access per agent, flow limits for agent-to-agent calls, and encrypted per-agent secrets. The control plane runs as a single process backed by Postgres, Redis and S3, and can be started with Docker Compose.
Key features
- 01Discovers installed coding agents such as Claude Code, Codex, OpenCode and Cursor CLI without changing their settings.
- 02Records coding-agent sessions and attributes token usage and cost per agent and per model.
- 03Routes harness LLM calls to a chosen provider while agents hold only short-lived tokens.
- 04Provides an MCP gateway that exposes only the tools each agent has been granted.
- 05Limits agent-to-agent call depth, fan-out and token budget to stop runaway loops.
- 06Stores per-agent secrets encrypted with AES-256-GCM.
Use cases
- Finding out which coding agents a team runs and what each one costs per month.
- Routing Claude Code or Codex traffic to a different model provider without changing developer workflows.
- Controlling which external tools an agent can call through a single MCP endpoint.
- Deploying and running A2A-compatible agents on a self-hosted control plane.
Quick start
- 1
Clone the repository and enter it
git clone https://github.com/Nasiko-Labs/nasiko.git && cd nasiko - 2
Install the CLI from this checkout (requires Rust)
cargo install --path cli/ --force - 3
Run the read-only discovery command to see which coding agents are running
nasiko agents discover
Things to know
- The CLI install path requires a Rust toolchain.
- Reporting and routing also need an active control plane and a valid login.
- The license metadata is unclear (NOASSERTION), though the README badge says Apache-2.0.
In collections
Topics
FAQ
- What is nasiko?
- An open runtime that discovers, reports on, and routes the coding agents already running on a developer's machine, with per-agent cost tracking. Nasiko is an open runtime for AI agents and coding harnesses. Many teams already run tools such as Claude Code, Codex, Cursor and OpenCode, each with its own billing and dashboard, which makes it hard to see who spent what and on which model. Nasiko detects these harnesses without modifying their configuration, and its CLI can record their session activity and attribute token usage and cost by agent and model.
- Is nasiko free to use? What license does it have?
- nasiko is open source under the Other license. Check the license terms before using it in commercial products.
- Is nasiko still maintained?
- Actively maintained. The last commit to nasiko was on Oct 23, 2026.
- What can I use nasiko for?
- Finding out which coding agents a team runs and what each one costs per month. Routing Claude Code or Codex traffic to a different model provider without changing developer workflows. Controlling which external tools an agent can call through a single MCP endpoint. Deploying and running A2A-compatible agents on a self-hosted control plane.
- How do I install nasiko?
- Clone the repository and enter it: git clone https://github.com/Nasiko-Labs/nasiko.git && cd nasiko Install the CLI from this checkout (requires Rust): cargo install --path cli/ --force Run the read-only discovery command to see which coding agents are running: nasiko agents discover
- Where can I get nasiko?
- The source code and installation instructions are on GitHub: https://github.com/Nasiko-Labs/nasiko
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
Self-hostable MIT AI gateway that exposes one OpenAI-compatible endpoint to hundreds of model providers, with quota-aware auto-fallback and token compression.
A single Rust binary that gives any AI SDK one endpoint to reach many LLM vendors, with MCP tools, traffic control and request logging.
Security controls that block, pause, and audit AI agent actions before they run, for Claude Code, OpenClaw, and MCP agents.
A token-efficient, microkernel AI agent for CLI, Web UI and chat channels that routes each turn to the cheapest capable model.
Self-hosted dashboard for dispatching tasks, reviewing agent runs, tracking spend, and coordinating AI agent runtimes like OpenClaw, Claude Code, and Codex.