Yes — free and open source under GPL-3.0. The code is on GitHub and contributions are welcome.

Does the AI send my data anywhere?

Only if you enable it, and only to the provider you chose with your own key — or nowhere at all, with a local model. The launcher is fully functional with AI off.

Which Linux desktops are supported?

All of them is the goal. Lychi runs on Wayland and X11, with dedicated backends where compositors need them. If yours misbehaves, open an issue.

Is this like Raycast or Spotlight?

Same reflex, different values: Linux-native, open source, and local-first.

Lychi — a fast, private launcher for Linux

Intelligence, on your terms

AI, on your terms. Including “no.”

Off by default. Bring your own key — or run a model locally if your machine can carry it. Nothing is required.

off · the default BYOK · key in your keyring local · Ollama or bundled, fully offline

Plain language, same deterministic commands.

With AI on, you stop memorizing prefixes. The agent works through the same command registry your fingers do — folded into a handful of typed tools, so every action it takes is one you could have typed. Risky steps ask first, and “Always allow” remembers your answer per action — never for anything high-risk.

It can search the web and read pages to answer with current information, and it can see: ask about your screen and it screenshots (hiding itself first), analyzes, and shows you the exact capture it looked at.

web search & fetch · consent-gated screenshot analysis documents & vision attachments

AI suggests. It never executes.

Every action — typed by you or proposed by a model — passes through one rules engine that owns shell, path, and URI decisions. Risky steps ask first. Fork bombs, curl | sh, rm -rf /, raw disk writes — denied outright, and a denial can't be confirmed away.

Commands resolve against your focused terminal or IDE — working directory, git branch, project type — including multi-repo workspaces. When the context is ambiguous, Lychi withholds it rather than guessing.

Privacy isn't a setting.

It's the architecture.

History, notes, clipboard, chats — local files with owner-only permissions. API keys in your system keyring, never on disk. Password-manager clipboard entries are skipped by default. Every feature that could touch the network is off until you say yes, and every grant is visible and revocable. No account, no telemetry.

web pages the AI reads are marked untrusted — data, not instructions corrupt data recovers to a timestamped .bak — never wiped shareable debug log: metadata only, secrets scrubbed AI chats auto-expire after 90 days

One file. One keystroke.

A single AppImage. No installer, no account, no telemetry.

Download AppImage

Then press Super+Space. Built for Linux desktops across the board — Wayland and X11 alike. Updates are manual for now; if something misbehaves on your setup, an issue report is very welcome.

Questions, answered

Is Lychi free?: Yes — free and open source under GPL-3.0. The code is on GitHub and contributions are welcome.
Does the AI send my data anywhere?: Only if you enable it, and only to the provider you chose with your own key — or nowhere at all, with a local model. The launcher is fully functional with AI off.
Which Linux desktops are supported?: All of them is the goal. Lychi runs on Wayland and X11, with dedicated backends where compositors need them. If yours misbehaves, open an issue — cross-desktop reliability is a priority, not an afterthought.
Is this like Raycast or Spotlight?: Same reflex, different values: Linux-native, open source, and local-first — your data never funds the product, because there's nothing to fund it with.