Back to directory

dsh-mesh

In a curated listMaintenance: Active

linyanzhi/dsh-mesh

Joins several DSH machines/phones into one mesh so agents can see peer nodes, read and transfer files across machines, and delegate tasks to another agent, with pairing codes and ACLs.

View on GitHubHomepage
$ dsh plugin --profile web add @linyanzhi/dsh-mesh

22

stars

4

forks

JavaScript

Language

MIT

License

2026-11-09

Created

2026-11-19

Last push

MIT, LAN-first multi-node network for DeepSeek Harness in early 0.x: one sidebar page for topology and node health, cross-machine file transfer and agent-to-agent delegation, held together by one-time pairing codes and per-peer ACLs, with no cloud relay.

DSH integration

Native runtime

Author-claimed

Safety audit

Unaudited

Last verified

2026-11-20

License

MIT

01What can it help you accomplish?

  • Keep every machine that runs DSH — office PC, home PC, phone — in one visible network

    A sidebar topology with all nodes, link type (LAN / public / tunnel), latency, version, instance state and offline channels marked red

    One developer’s own 2+ always-on machines; the README is explicit that this is not a team or multi-user tool

  • Move work between machines instead of moving files by hand

    Cross-machine file reads, chunked mesh transfer with size and hash verification, and delegation records (task envelope + state machine + result written to disk on the peer)

    Developers who want the agents to hand tasks over and check the result themselves

  • Roll a plugin install or a DSH upgrade across every device from one place

    One delegation that has the other machines install, upgrade and verify, with each node keeping its own state and logs

    People maintaining several DSH installs who want them pinned to the same version

02How to install into DeepSeek Harness

Prerequisites

  • Each machine: Windows 10/11 x64 is the primary target; macOS / Linux can run the server and bridge, but autostart and similar features were not verified
  • DSH `0.2.2-rc.1` (compatible with `0.2.0-rc.2`–`0.2.x`); keep every machine on the same version where possible
  • Node `22.19+` or `24+` (the runtime DSH already bundles)
  • Admin rights — the plugin install, firewall rules and autostart all touch the system
  • Network: a shared LAN where the machines reach each other on `8124/tcp`, or a public TCP ingress you control (self-hosted frps, SakuraFrp, Cloudflare Tunnel, Tailscale)

Installation steps

  1. 01

    Run `dsh plugin --profile web add @linyanzhi/dsh-mesh` on every machine — the single package brings the UI page, the mesh server, the host bridge and the autostart guard

    $ dsh plugin --profile web add @linyanzhi/dsh-mesh

  2. 02

    Restart that DSH instance (refreshing the page is not enough); the “AI 网络” entry appears in the left sidebar

  3. 03

    On the machine to be reached: AI 网络 → 局域网连接 → generate my pairing code (one-time, 10 minutes by default)

  4. 04

    On the other machine: scan the LAN and enter the code, or type `https://IP:8124` plus the code, then pair with that node

  5. 05

    For a cross-network link, change the listen address to `0.0.0.0:8124` first and map `domain:port` through your frp or tunnel

Verify the integration

  • `dsh-mesh host status` — whether the host bridge is alive inside DSH
  • `dsh-mesh cluster status` — instances and latency for this machine plus every paired node
  • Or ask the agent to look at the AI network status and list all nodes

03DSH integration and capability boundaries

DSH integrationNative runtime

Ships as a single DSH plugin: `dsh plugin --profile web add @linyanzhi/dsh-mesh` installs the “AI 网络” sidebar page, one host bridge per DSH instance and one mesh server (MCP-over-HTTPS) per node, so peer DSH agents can list nodes, read files and delegate tasks to each other.

  • Sidebar topology (“AI 网络”)

    the paired nodes on the mesh→Topology view with link types, per-channel address and status, offline markers, and one host bridge listed per DSH instance

    Public IP / domain is masked as “公网地址已隐藏” by default and only revealed through a header toggle
  • Cross-machine file transfer

    a file path on a peer node→Chunked transfer with size and hash verification, plus transfer records for this machine and for the whole network

  • Agent-to-agent delegation

    a task envelope sent from one DSH to another→A state machine per task, the result written to disk on the peer, and a delegation ledger in the panel

    The delegated turn runs on the peer with that peer’s own local tool surface
  • Host bridge

    a running DSH instance→Lets the local mesh server create sessions, post messages and call local tools inside that instance

    One bridge per DSH instance, listening on 127.0.0.1 only (9100+ by default)
  • Pairing codes and per-peer tokens

    a one-time pairing code generated by the receiving side→A per-peer token that the server stores only as a hash and that can be revoked at any time; tokens live in each node’s `~/.dsh-mesh/`

    Discovery is not connection: a LAN scan only lists nodes, and nothing connects until a code is redeemed

04Who is it for? When not to use it?

Good for

  • One developer’s own 2+ always-on machines; the README is explicit that this is not a team or multi-user tool
  • Developers who want the agents to hand tasks over and check the result themselves
  • People maintaining several DSH installs who want them pinned to the same version

Not for

  • 0.x preview: interfaces and the config format still change, so read the changelog before upgrading.
  • Windows 10/11 x64 is the tested target. macOS and Linux can run the server and bridge, but autostart and remote restart were only built for Windows, and phone nodes (Android + proot) are experimental.
  • Every node needs a reachable TCP `8124` — the same LAN, or an ingress you run yourself (frps / SakuraFrp / Tailscale / Cloudflare Tunnel). There is no cloud relay: if the tunnel or the LAN is down, the peer is simply unreachable.
  • The default ACL allows read/search/write inside the peer user’s `~/Code`, reading session history and logs, and delegating tasks; `shell_run`, channel messages, remote desktop and repair actions stay off until opened, and desktop control (`pc_action` / `pc_app` / `pc_window` / `pc_browser`) is opt-in. `/api/dsh-mesh/*` only answers loopback requests, and there are no team or multi-user permissions.

05Compatibility, maintenance and safety notes

  • 0.x preview: interfaces and the config format still change, so read the changelog before upgrading.
  • Windows 10/11 x64 is the tested target. macOS and Linux can run the server and bridge, but autostart and remote restart were only built for Windows, and phone nodes (Android + proot) are experimental.
  • Every node needs a reachable TCP `8124` — the same LAN, or an ingress you run yourself (frps / SakuraFrp / Tailscale / Cloudflare Tunnel). There is no cloud relay: if the tunnel or the LAN is down, the peer is simply unreachable.
  • The default ACL allows read/search/write inside the peer user’s `~/Code`, reading session history and logs, and delegating tasks; `shell_run`, channel messages, remote desktop and repair actions stay off until opened, and desktop control (`pc_action` / `pc_app` / `pc_window` / `pc_browser`) is opt-in. `/api/dsh-mesh/*` only answers loopback requests, and there are no team or multi-user permissions.
2026-11-092026-11-17Not specified by the author

MIT · actively developed, no GitHub release yet (last push 2026-11-17) · early 0.x preview

06Frequently asked questions

How many machines do I need for dsh-mesh to be useful?

At least two machines you keep powered on that both run DSH, plus the willingness to install from the command line and open a port or set up a tunnel. With one machine you only ever see yourself, and the README lists that case under “not for you”.

Does my traffic go through a third-party cloud?

No. The design is LAN-first, and when you cross networks you bring your own relay — self-hosted frps, SakuraFrp, Tailscale or Cloudflare Tunnel. Data stays between your machines and your own tunnel; there is no cloud account in the loop.

What can a paired peer do on my machine by default?

The default ACL allows reading, searching and writing inside the peer user’s `~/Code`, reading session history and logs, and delegating tasks. `shell_run`, channel messages, remote desktop and repair-type operations are off by default, and desktop control (`pc_action` / `pc_app` / `pc_window` / `pc_browser`) is opt-in. The `/api/dsh-mesh/*` endpoints only accept loopback requests.

Is dsh-mesh a remote desktop or SSH replacement?

No, and the README says so explicitly: desktop control is an optional capability that ships off. The point of the plugin is agents calling each other — you pair once and watch the status, the machines do the moving.

How do I confirm the install and the pairing worked?

`dsh-mesh host status` reports whether the host bridge is alive inside DSH, and `dsh-mesh cluster status` lists instances and latency for this machine plus every paired node. You can also just ask the agent to look at the AI network status and list all nodes; the “AI 网络” sidebar entry is the UI side of that same check.

08Data and sources

  • Author-claimedgithub.comff3158d2bc73…

    A multi-node AI network for [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness): turn your PCs and phone…

  • Author-claimedgithub.comff3158d2bc73…

    > 一条命令装齐(推荐):UI 页面 + mesh 服务端 + 宿主桥 + 自启守护

  • Author-claimedgithub.comff3158d2bc73…

    - **host bridge**:DSH 宿主插件,让 mesh server 能创建会话、发消息、调用本机工具

This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-11-20. Found an error? Submit a correction.

🏆

Best DeepSeek Harness Plugins

Twelve plugins worth installing first — picked from the whole catalog, across every category.

DSH Plugins is an independent community directory of DeepSeek Harness plugins. Not affiliated with or endorsed by DeepSeek. Third-party plugins are not security-audited — review the source before installing.

New DeepSeek Harness plugins, weekly. No spam.