Langoliers

Early release · scans are read-only · actions are dry runs

It follows behind you, and does nothing until you answer.

Langoliers is a local agent that finds the work you can't get to — tickets assigned to you that you never touched, threads that need a reply, reviews waiting on you — then asks what to do about each one, as multiple choice with a recommended default.

Install on macOS

curl -fsSL https://get.langoliers.enginyyr.com/install.sh | sh

Verifies a SHA-256 before installing. Puts langoliers and langoliersd in ~/.local/bin. No sudo. Details and uninstall

The terminal interface

The Langoliers terminal interface. A list on the left shows three items under Needs you: an export job that times out, a thread that mentions you, and a refactor waiting on your review. The card on the right shows the first item with three numbered options and an Other option. Option 1, Ask for reproduction steps, is marked recommended and shows a dry run of the comment it would post. A banner reads DRY RUN: nothing is written.

The real interface, rendered from the project's snapshot tests with invented data. Number keys answer; Enter takes the recommendation.

The backlog nobody is looking at

More is assigned to you than you can handle. You focus on a few things each day and the rest quietly rots. Nothing is on fire, so nothing surfaces, and the mess grows until it does.

Trackers show you everything and prioritize nothing. Always-on cloud agents offer to handle it for you, but you may not want an agent acting for you inside work systems, and you may not be allowed to connect one. Langoliers works behind you, not instead of you.

The name comes from Stephen King's The Langoliers: creatures that devour whatever has already passed. Here they eat the stale backlog behind you.

Find, ask, act, remember

  1. Find

    Scans read open items assigned to you, your mentions and your requested reviews across the systems you connect.

    Works today: GitHub and Linear, read-only

  2. Ask

    Each item becomes a card: a short summary and numbered options with a recommended default. You can also write your own instruction.

    Works today

  3. Act

    Only after you answer, and only with what that option needs. Today the answer is recorded and the action is shown as a dry run.

    Dry run only: nothing is written yet

  4. Remember

    Repeated answers become rules, so each scan is quieter than the last.

    Not built yet

Approval-first, and it stays on your machine

What it isn't

Where it stands

Langoliers is early. This is what exists and what doesn't, so you don't have to guess.

Works today

  • A Rust daemon and a terminal interface, with a demo mode that runs entirely on fake data
  • Read-only scans of GitHub and Linear for a personal account
  • Triage by Claude Code, run with no tools, with output validated and rejected if malformed
  • Cards with multiple-choice options, a recommended default, notes and free-text instructions
  • Recorded decisions, a history per card, and a dry run that shows what would have been done

Not yet

  • Executing approved actions: execution is off, so nothing is written to GitHub or Linear
  • Rules learned from your answers
  • Codex as an agent; only Claude Code is wired up
  • Scheduled scans: a scan runs when you press s
  • Linux, Windows, or a graphical interface

Install

macOS, Apple silicon or Intel

curl -fsSL https://get.langoliers.enginyyr.com/install.sh | sh

What the installer does

  • Downloads the release archive over HTTPS and checks its SHA-256 against a manifest before unpacking anything.
  • Installs langoliers and langoliersd into ~/.local/bin, or the folder in LANGOLIERS_INSTALL_DIR. No sudo, and nothing outside that folder.
  • Replaces only files it installed itself, so running it again upgrades in place.
  • The whole script is one function that runs last, so a cut-off download does nothing.

About signing

The binaries are not Developer ID signed or notarized. A file fetched with curl isn't quarantined, so macOS doesn't prompt, but that also means no Apple review stands behind it. The checksum catches a corrupted or truncated download. It can't vouch for the host, because the manifest is served from the same place as the archive. If that matters to you, read the script first: curl -fsSL https://get.langoliers.enginyyr.com/install.sh | less.

Then

langoliers --demo               # fake data; starts langoliersd, needs no agent CLI or credentials
langoliersd connect github      # personal account, read-only
langoliers                      # press s to scan

To remove it: curl -fsSL https://get.langoliers.enginyyr.com/install.sh | sh -s -- --uninstall. Your data in ~/.langoliers is left alone.