SOFTWARE / TOOLS / EXPERIMENTS

A small toolkit.
Connected work.

Projects for machine coordination, model routing, and coding-agent workflows. Explore what each tool does, how it fits, and its current development stage.

BY Y.D. SUN BEIJING

THE TOOLKIT / 001

The projects.

A closer look at the tools and their development.

01 / CONNECT YOUR MACHINES

tailnet-mesh

Self-hosted

Turn your Tailscale-connected machines into a coordinated workspace for computing, collaboration, and shared services.

  • Mesh networking
  • Distributed compute
  • Operations console

A CLOSER LOOK

Your machines, working together.

A unified Python toolkit for the machines on your tailnet. Manage nodes from a terminal or web console, distribute compute tasks, and bring shared services into one operational view.

For developers and researchers managing their own multi-machine environment.

Research & development project
  1. See the whole fleet

    A unified terminal and web operations console for nodes, jobs, runtimes, storage, and network services.

  2. Put spare compute to work

    SSH-based task distribution across Windows, Linux, and macOS, with resource-aware placement.

  3. Share services on your terms

    Self-hosted Git, file sharing, collaboration, and reviewed operational workflows over Tailscale.

Development stageSelf-hosted on your own Tailscale network. Setup and platform checks are described in the project documentation.

02 / CONNECT YOUR AI TOOLS

AI Agent Gateway

Desktop preview

A local AI gateway and configuration manager. Bring coding tools and model providers together with explicit routing and fallback.

  • AI Agent Gateway
  • Provider routing
  • Local configuration

A CLOSER LOOK

One local gateway. More choice.

Also called AI Agent Gateway (AAG), cc helps you manage providers, connect AI coding tools, and configure ordered fallback routes through a local gateway.

For developers working with multiple coding tools and model providers.

Explore the repository
  1. Connect compatible tools

    OpenAI- and Anthropic-compatible endpoints, with Responses API support in the desktop Rust proxy.

  2. Keep routing explicit

    Ordered provider fallback for retriable failures, with per-provider circuit breakers.

  3. Manage configuration locally

    A desktop interface for providers, routes, integrations, backups, and configuration; a headless Node gateway can run independently.

Development stageDesktop v0.2.0 preview, with an independently runnable Node gateway. See the public repository for setup and release notes.

03 / COORDINATE YOUR WORK

vibemux

Pre-alpha

A Windows-first environment being built for coding-agent workflows, isolated Git worktrees, and an auditable record of what happened.

  • Agent workflows
  • Git worktrees
  • Local-first

A CLOSER LOOK

Give agent work a place to live.

VibeMux is a developing, local-first coordination environment. It explores how worktrees, a Rust daemon, and explicit review can make coding-agent workflows easier to inspect.

For developers exploring local, inspectable agent coordination.

Explore the repository
  1. Separate changes by worktree

    A separate Git worktree for each run, with reviewable diffs and deliberate cleanup.

  2. Keep a durable record

    A local Rust daemon with authenticated control, lifecycle inspection, state, events, and explicit recovery.

  3. Build review into the loop

    Bounded local planner, worker, and reviewer workflows, with independent review and local verification required for completion.

Development stagePre-alpha. Native multi-harness integration is the next development focus; local workflows and the Rust daemon form the current foundation.

THE COMMON THREAD / 002

Working principles.

Understandable tools, clear ownership, and work you can inspect.

Infrastructure, intelligence, and workspace are different layers. These projects explore each one with a shared preference for local control, clear ownership, and reviewable work.

01

Independent by design

Each project keeps its own purpose, setup, and development path.

02

Control stays explicit

Providers, machines, credentials, and execution need deliberate configuration.

03

Progress you can inspect

Readable source, reproducible checks, and a clear account of the current development stage.

GOOD TO KNOW / 003

Practical details.

Setup, scope, and getting started.

01How do the projects fit together?

Tailnet Mesh coordinates machines, AI Agent Gateway routes model requests, and VibeMux organises coding work. Each project has its own setup and can be explored independently.

02What is ready to use today?

AI Agent Gateway provides a desktop preview and a standalone Node gateway. VibeMux is in pre-alpha. Tailnet Mesh is a self-hosted toolkit for a configured Tailscale network. The repository pages contain current setup instructions.

03Where do model requests run?

Routing is configured locally. Requests go to the model providers and endpoints you choose, while Tailnet Mesh coordinates the machines on your own network.

04Where can I get started?

Open the public GitHub repositories for installation instructions, source code, and development notes. For Tailnet Mesh, contact me to discuss the project and its deployment model.

SOURCE & DEVELOPMENT

Follow the work.

Explore GitHub