Show HN: Ts-SSH – SSH over Tailscale without running the daemon

ts-ssh solves a specific problem: accessing machines on your Tailnet from environments where you can't install the full Tailscale daemon (like CI/CD runners or restricted systems).

  It uses Tailscale's tsnet library to establish userspace connectivity, then provides
  a standard SSH experience. Works with existing workflows since it supports normal SSH
   features like ProxyCommand, key auth, and terminal handling.

  Some features that proved useful:
  • Parallel command execution across multiple hosts
  • Built-in tmux session management for multi-host work
  • SCP-style file transfers
  • Works on Linux/macOS/Windows (AMD64 and ARM64)

  The codebase is interesting from a development perspective - it was written almost
  entirely using AI tools (mainly Claude Code, with some OpenAI and Jules). Not as an
  experiment, but because it actually worked well for this kind of systems programming.
   Happy to discuss the workflow if anyone's curious about that aspect.

  Source and binaries are on GitHub. Would appreciate feedback from anyone dealing with
   similar connectivity challenges.
github.com

・

103 points

・

i8code

・

13 days ago

42 comments

anotherpaulg ・ 12 days ago

Looks very interesting. I was hoping it would solve a problem I’ve had recently:

I want to ssh into a windows box that I only have a normal user account on. So I can’t (and don’t want to) change any admin settings or install anything as admin.

All the obvious approaches hit roadblocks.

Seems like this tool solves the opposite problem: sshing out from a minimally privledged environment.

  • paxys ・ 12 days ago

    You can start your own ssh daemon from the unprivileged account pointing to a random port.

    • anotherpaulg ・ 12 days ago

      Ya, you would think so. But when you connect to it and sshd tries to fork a process to handle the session… you get a privileges error.

un1970ix ・ 12 days ago

This is why you don't let Claude handle versioning and the release process. From v0.1.0 to v1.0.0 to v2.0.0, and then suddenly 1.2.0? Semantic versioning isn't quantum mechanics. (Even then, I'll admit it's sometimes hard for me too to decide the right increment when tagging versions. :)

  • KomoD ・ 12 days ago

    > This is why you don't let Claude handle versioning and the release process.

    Or anything else without reviewing it.

    lol @ the issue in the repo: "module declares its path as: github.com/yourusername/ts-ssh"

  • ape4 ・ 12 days ago

    In general, maybe security code (which is what this is) might not be the best place for AI.

  • cedws ・ 12 days ago

    Semantic versioning isn’t the only way of versioning. Linus Torvalds versions Linux how he sees fit.

  • ramon156 ・ 12 days ago

    - 0.0.1 -> improvements

    - 0.1.0 -> breaking changes

    - 1.0.0 -> overhaul/refactor needed

    I know not every case is easy but this is my rule of thumb. I've honestly never needed a major version change

    • Timon3 ・ 12 days ago

      That kind of goes counter to semantic versioning, where:

      - x.y.Z (patch) -> backward compatible bug fixes

      - x.Y.z (minor) -> backward compatible new features

      - X.y.z (major) -> breaking changes

      But of course it's fine to use whatever versioning scheme you like, as long as you communicate it to your consumers.

      https://semver.org/

      Edit: updated the version strings for clarity.

    • chrisweekly ・ 12 days ago

      Using a 3-digit version like semver, while assigning different semantics, is a recipe for confusion if anyone except you ever refers to a package using this "rule of thumb".

    • indrora ・ 12 days ago

      0.0.1 - I fixed a bug

      0.1.0 - I rearchitected the bug

      1.0.0 - The bug is integral to the codebase.

      • i8code ・ 11 days ago

        I agree, the initial set of releases were all over the place. I took the feedback from this thread and fed it to Claude along with the semver.org references that were linked here for more detailed (and pedantic) context. Makes way more sense now. Thanks for the feedback! Claude handled the cleanup. Here's the updated releases: https://github.com/derekg/ts-ssh/releases

huslage ・ 12 days ago

I am scared that this is vibe coded and not audited in any way. tsnet is good software, but wrapping it in this way is a recipe for disaster. Please reconsider.

  • isatty ・ 12 days ago

    I agree and had the same thought. Tailscale ssh is good and I was interested in something like this but absolutely not if it’s AI generated garbage.

  • mystifyingpoi ・ 12 days ago

    > I am scared that this is vibe coded

    Totally serious question: would you feel better about this piece of software, if you didn't know that it was vibe coded?

    Do we need "build without AI" stickers on every piece of software created these days?

    • huslage ・ 12 days ago

      I looked at the code and the documentation and it's definitely vibe coded. Also the presence of CLAUDE.md is pretty telling. I have no issue with vibe coding in general, but I am skeptical of the usefulness of LLMs with security code.

      Yes, I think projects that are coded wholly or in part by LLMs should be noted as such.

      • eddd-ddde ・ 12 days ago

        Why would you trust a random person's project anymore than an AI project? I'd say the vast majority of the population is vastly less skilled than Claude Code.

        I.e. just because it's human doesn't mean it's any more secure.

    • cchance ・ 12 days ago

      [flagged]

      • KomoD ・ 12 days ago

        I don't really care if "AI assistance" was used as long a human is actually reviewing the output, which just doesn't seem to be the case here (and usually not the case when it comes to "vibe coding")

      • eknkc ・ 12 days ago
        ・ 2 more

        I feel fine if AI was used to add features to an established software. Let it loose on the linux kernel for what I care. It still somehow feels icky to use it to build something from scratch.

        • elashri ・ 12 days ago

          Ironically it wouldn't be very useful for Linux kernel development (would be very hard to out it in context) while it is more suitable for new projects written from scratch.

          This of course not considering the quality or anything else.

      • Bender ・ 12 days ago
        ・ 2 more

        Somewhat off topic question but I ask this from time to time and maybe now is that time. Has AI started fixing everyone's software bugs and closing out all the CVE's yet?

        • cchance ・ 7 days ago

          I saw a youtube video of a few CVE's being closed out by some automated AI bot they were talking about recently i can't remember the channel sorry :(

          That said coderabbit's pretty damn impressive for just generally reviewing PR's and catching shit for human review.

      • un1970ix ・ 12 days ago

        No one is against using AI or coding with agents unless you don't understand what it's doing and you're incapable of reviewing the output. The problem isn't the tool, it's "coders" who unthinkingly trust it without verification.

  • KetoManx64 ・ 12 days ago

    Can you explain what the possible risks are?

amacneil ・ 12 days ago

Meta question: How does a HN post like this come to exist with _both_ a link and a body?

Anytime I've submitted with both url + body the body is posted as a comment.

  • pvg ・ 12 days ago

    Show HN's get to have text and a link, most other things don't.

  • ambigious7777 ・ 12 days ago

    not sure, but i think this may be a special feature for Show HNs

rsync ・ 12 days ago

Tangential ... I think I read somewhere that I cannot become a customer of tailscale without FAANG credentials ?

As in, I cannot simply sign up with my own personal identifiers (email, phone, etc.) but need to participate in a google auth or FB auth mechanism ?

I found it hard to believe - is this, indeed, the case ?

  • _ks3e ・ 12 days ago

    It's possible to use Tailscale with just a passkey [0], but it's a weird process because they don't let you create a tailnet and a passkey account at the same time. Instead, you need to create an account with a throwaway FAANG credential and send yourself an invite to that account's tailnet, and then use that invite to create a passkey-linked Tailscale account. This account can then create its own tailnet, at which point the original tailnet (and the throwaway FAANG account) can be discarded.

    It's a weird process and not particularly user friendly (passkey accounts are tied to a specific passkey and can't have additional ones added, so you need to create a new account if you, say, migrate from one hardware key to another). Hopefully they improve the process before passkey support goes out of beta.

    [0] https://tailscale.com/kb/1269/passkeys

    • xeonmc ・ 12 days ago

      I feel like maybe they should allow adding SSH keys as a login method instead of passkeys.

      Though I suppose there is the potential problem of identitiy collision due to public key resuse unless the keys were generated serverside to guarantee uniqueness.

  • erinnh ・ 12 days ago

    You need one of the following:

    Google, Microsoft, Github, Apple or your own OIDC Provider.

    They do not have their own account backend.

    So you dont technically need a FAANG account if you have a Gitea, Gitlab, Authentik Account or something like that.

  • seized ・ 12 days ago

    You can also use Codeberg.