Thread Nimbus is now free: an SSH client that lets an AI help, without letting it loose

Thread Nimbus is free to download. Real terminals, OS-keychain credentials, verified host keys, and an AI agent that cannot run a destructive command even if it wants to.

Thread Nimbus terminal session

Thread Nimbus is now free. No card, no trial timer: give an email and a phone number and the install command appears immediately. Download it at nimbus.threadlane.in.

This post explains what it actually does, and the specific problems it was built to solve.

The problem with putting an AI in your terminal

Most “AI terminal” tools work by giving a language model a shell and hoping. That is fine on a laptop and unacceptable on a production host. The failure mode is not subtle: a model that confidently runs rm -rf against the wrong path, or pastes a private key into a prompt that gets logged somewhere you do not control.

Thread Nimbus starts from the opposite assumption. The agent is useful precisely because it cannot freestyle. Every capability below exists to make an AI-assisted session something you could show an auditor.

Real terminals, not a chat box pretending

Nimbus runs genuine SSH sessions — ssh2 underneath, xterm.js on screen. Your prompt, your shell, your dotfiles, your TUI apps. Twenty-one terminal themes ship with it, switchable live, four of them free and the rest a one-time two dollars.

This matters more than it sounds. A tool that emulates a terminal instead of running one breaks the moment you open htop, resize a pane, or use anything with cursor addressing.

Secrets that never touch disk

Passwords and key passphrases live in your operating system keychain. The app stores keytar references — never the secret itself — and never sends them to a model. Host keys are verified, so a machine-in-the-middle on a hostile network fails closed instead of silently succeeding.

If you have ever inherited a team SSH config with plaintext passwords in it, this is the part worth caring about.

You approve every change

This is the core of it. Every command the agent proposes is classified in the main process, before execution:

  • Read-only commands can run automatically.
  • Mutating commands require your explicit confirmation.
  • Destructive patterns — rm -rf /, mkfs and friends — are blocked outright and never offered for confirmation at all.

The classification happens in the main process, not the renderer, which means the UI cannot bypass the gate even if it is compromised. That is the difference between a safety feature and a safety theatre.

The features you end up using daily

Command Audit Trail — every command, who ran it, when, and what came back. Exportable, which turns “what happened at 03:00?” into a query instead of an argument.

Secret Shield — redacts credentials and tokens from the screen, so a scrollback does not become a leak.

Presentation Mode — one shortcut and your terminal is safe to share on a call: secrets masked, hostnames blurred. Anyone who has demoed against production knows why this exists.

Multi-Host Broadcast — run the same command across a fleet, with the same guardrails applied per host.

Parameterized Runbooks — the operational knowledge that normally rots in a wiki, executable and reviewable.

Devi Assistant, if you want it

Devi is the AI agent that works over your live authenticated session — investigating, explaining, and proposing fixes with the full context of the box you are actually on. It is off by default and optional at fifteen dollars a month, or a hundred and forty-four a year.

The app is completely usable without it. That is deliberate: the SSH client should not be held hostage to a subscription, and inference costs real money, so it is priced separately rather than bundled into everything.

Installing it

Two ways, both starting at nimbus.threadlane.in:

  1. Give your email and phone, then use the download link for macOS (.dmg), Windows (.exe) or Linux (.AppImage).
  2. Or install from a terminal with the one-line command shown after you submit.

Your email is your licence. Come back to your account any time to re-download it — new laptop, fresh machine, years later. The link is permanent and available once a week.

The rest of what we build

Thread Nimbus is one of several products from ThreadLane, where we build enterprise-grade AI systems — and build them for other companies too, from an idea to a production pilot in a median of six weeks.

  • ThreadLion Code — a CLI coding agent that maps your repository and passes eight gates proving it understands the code before it edits a single file. Same philosophy as Nimbus, applied to source control.
  • ThreadOps — 49 AI agents running your marketing: SEO articles, social posts and ad creative, with one-tap approval on Telegram. Live in about two hours, from $89 a month, no extra headcount.
  • ThreadRank — check whether ChatGPT, Gemini, Claude and Google can actually cite your site. Free instant scan, separate SEO and AEO scores out of 100.
  • ThreadLane — the portfolio and community platform for fashion designers and textile creators, bridging traditional craftsmanship with the tools to build a design career.

If you are a larger team and want this kind of system built for you rather than bought off the shelf — AI agents, RAG pipelines, LLM fine-tuning, or the infrastructure underneath them — that is exactly what ai.threadlane.in is for. We build enterprise AI products for big tech, and the products on this page are what that work looks like when we ship it for ourselves.

Start here

Thread Nimbus is free. Get it at nimbus.threadlane.in, and tell us what breaks.