dsh-devices

English · 简体中文

Turn your devices into a fleet.

Zero core changes, pure plugin mounting. A dsh plugin that lets your devices collaborate. Tools auto-register in dsh sessions — any dsh-hosted agent can call them directly. Uninstall leaves no core patches.

Requirements: ① At least 2 devices, each with dsh-devices installed (pairing is two-sided; a single device has nothing to pair with). ② Connectivity: same LAN with mDNS multicast allowed (UDP 5353; doesn't cross routers/subnets), or publicly reachable devices with SSH (port 22) open. NAT-isolated devices cannot connect to each other in this version (V2 = P2P/tailscale planned).

┌──────────────┐          ┌──────────────┐
│  dsh + devices  │◄────────►│  dsh + devices  │
└──────────────┘          └──────────────┘

npm install -g @deepseek-ai/dsh

dsh plugin add dsh-devices

Platform Notes

Verified on real hardware: macOS / Ubuntu 22.04+ / Windows Server 2025 & Windows 11.

macOS

• mDNS auto-announce starts with the plugin (no manual fleet7 serve); if discovery fails, allow UDP 5353 in the firewall.
• Remote login (SSH server) is off by default — enable it in System Settings → General → Sharing → Remote Login for other devices to command this machine.
• Fleet keys live in ~/.fleet/ssh-keys/ (0600 enforced).

Linux (incl. Ubuntu / cloud servers)

• Node ≥ 22.19 required (22.14 and below lack the zstd API → createZstdDecompress boot error). Upgrade via nvm or the distro.
• As a managed host: sshd running, authorized_keys 600, ~/.ssh 700.
• Cloud security groups must allow SSH (22), or public pairing will time out.
• Slow networks: use npm config set registry https://registry.npmmirror.com to speed up dsh installs.

Windows (10/11 desktop and Server 2016+)

• Node ≥ 22.19 required — the zip build is most reliable (MSI silent install can silently fail on low-memory VMs).
• As a managed host: install OpenSSH Server (Settings → Optional features), then:

  Start-Service sshd; Set-Service sshd -StartupType Automatic
  

  Allow port 22 through the firewall.
• As a controller: single commands, fleet8 ssh, upload/download all work; && chains may return only the first command's output (OpenSSH for Windows console limitation — see TROUBLESHOOTING Q7). Split compound commands.
• fleet7/fleet8 exit cleanly on Windows (pipe-handle handling is explicit).

Known Limitations

• mDNS is LAN-only (beta): discovery works on one subnet only; cross-network devices use the SSH route.
• mDNS auto-announce requires the plugin loaded in dsh (it starts automatically).
• Windows as a controller: fleet_ssh_exec with && chains may return only the first command's output (OpenSSH for Windows console-session quirk; see TROUBLESHOOTING Q7). Windows as a managed device is unaffected.

Tools

Disabled modules don't register their tools (modules: mdns | ssh | both).

No commands to memorize — discovery, pairing, and execution are all done by the agent calling fleet_discover / fleet_pair / fleet_ssh_exec in your session.

Or grab the tgz from Releases: dsh plugin add ./dsh-devices-<version>.tgz.

dsh plugin add dsh-devices

dsh plugin add ./dsh-devices-<version>.tgz

You should see a clean install with no warnings. Four fleet_* tools are then auto-registered in dsh sessions.

git clone https://github.com/polaris-smart/dsh-devices.git

Scenario A: Same LAN (recommended start)

Both devices need dsh-devices (pairing is two-sided). Devices A and B on the same LAN.

B's user asks B's dsh agent to "start fleet broadcast" (agent runs fleet7 serve), or manually:

fleet7 serve

B generates a device key (fleet-d-..., printed in the terminal) on first fleet7 serve. Then on A:

Pair once, use forever — B has authorized A's public key; A remembers how to reach B.

Scenario B: Cross-network (SSH direct)

When devices are not on the same LAN (e.g., commanding a cloud server from home):

echo "ssh-ed25519 AAAA... my-server" >> ~/.ssh/authorized_keys

fleet8 ssh my-server "hostname"

Then use fleet_ssh_exec / fleet_workspace in dsh sessions.

Scenario C: One-command fleet join (new in v0.1.2)

Onboarding a new device in three commands — the invite code travels over WeChat/phone, no manual key setup:

Host — generate an invite code:

fleet8 invite <your-ip-or-host> --user <remote-user> --port 22
# prints an invite code: host:port/user/token — send it to your teammate

New device — join with one command:

fleet8 join <invite-code>
# parses the code, generates an SSH keypair, registers the device, prints the public key

Host — authorize:

fleet8 allow <device-id> "<public key from the step above>"
# done — the new device can now be reached via fleet8 ssh / fleet_* tools

|---|---|---|---|

Disabled modules don't register their tools.

|---|---|---|

Errors carry next-step guidance; file transfer & workspace go through the agent tools.

After install, just say "call fleet_discover" in your dsh session to see LAN devices; pair once, then command remote devices with fleet_ssh_exec. No extra config — tools are exposed to every agent in the dsh session.

Only paired devices are reachable; key files must be 0600; all errors return readable text instead of throwing.

|---|---|---| | modules | both | mdns / ssh / both |

┌─────────────────────────────────────────────────────┐
│                     dsh (DeepSeek Harness)                        │
│  ┌──────────────┐  ┌──────────────┐  ┌───────────┐  │
│  │ fleet_pair    │  │ fleet_workspace│  │           │  │
│  └──────┬───────┘  └──────┬───────┘  └───────────┘  │
│         │ mDNS            │ SSH                     │
└─────────┼─────────────────┼─────────────────────────┘
          ▼                 ▼
   ┌────────────┐    ┌──────────────┐
   └────────────┘    └──────────────┘
          │                 │

For full troubleshooting see TROUBLESHOOTING.md.

A: Both devices must be on the same LAN (mDNS multicast doesn't cross routers); the target must run fleet7 serve; allow UDP 5353 through firewalls.

A: Use Scenario B manual pairing; check port 22 reachability and that the public key is in authorized_keys.

A: You likely have an outdated package — download the latest tgz from the Releases page.

A: After restarting the dsh session, type /fleet — seeing your device identity and paired devices means it works. Or ask the agent to call fleet_discover.

A: Check: ① the plugin was added to the profile you're booting (dsh plugin --profile <name> add dsh-devices) ② the session was restarted ③ Node 22+ (node --version). Still missing? Open an issue with the dsh plugin output.

A: Verify port 22 reachability (nc -vz <ip> 22 or fleet8 ssh <name> hostname) and that the public key is in the target's authorized_keys. Timeouts are usually network/firewall issues, not plugin issues.

Found a bug or have an idea?

  Task-board claiming: post tasks from any device, paired devices claim them

Changelog

v0.1.1 (2026-08-20)

• 🔒 Internal codenames removed from the package (mDNS service type & SSH key comment now dsh-devices); per-platform notes added to README.

v0.1.0 (2026-08-20)

• 🎉 First release on npm + GitHub: dsh plugin add dsh-devices
• ✨ mDNS auto-announce (no manual serve), fleet_status liveness probe (7th tool), real /fleet commands, SFTP transfer, SSH connect-retry, Windows adaptations
• 🧪 92 tests + verified on macOS / Ubuntu / Windows Server 2025

History (pre-rename, v0.2.x era)

• v0.2.10 → v0.2.0: /fleet command, SFTP, zero-dependency rebuild, CLI precompile, decentralized pivot (mDNS + SSH)

MIT © polaris-smart