Show HN: Ts-SSH – SSH over Tailscale without running the daemon
github.comts-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.
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.
You can start your own ssh daemon from the unprivileged account pointing to a random port.
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. :)
> 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"
In general, maybe security code (which is what this is) might not be the best place for AI.
https://preview.redd.it/ou8h9owirmme1.png?width=1024&auto=we...
Semantic versioning isn’t the only way of versioning. Linus Torvalds versions Linux how he sees fit.
- 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
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.
Minor note, but since the topic is accurate semvar: rule 4 specifies 0.x.y is a special case where anything may change at any time.
Relevant section: https://semver.org/#:~:text=How%20should%20I,next%20major%20...
Good point, if we're already being pedantic we should be accurate. I've updated the version strings in my comment, thanks!
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".
0.0.1 - I fixed a bug
0.1.0 - I rearchitected the bug
1.0.0 - The bug is integral to the codebase.
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.
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.
> 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?
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.
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.
[flagged]
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")
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.
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.
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?
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.
Can you explain what the possible risks are?
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.
Show HN's get to have text and a link, most other things don't.
not sure, but i think this may be a special feature for Show HNs
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 ?
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
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.
Since April 2023 they support custom OIDC providers[1], and as of April 2024 that was extended to the free plan as well[2], so you can bring your own auth.
[1]: https://tailscale.com/kb/1240/sso-custom-oidc
[2]: https://tailscale.com/blog/sso-tax-cut
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.
You can also use Codeberg.