contents · Getting started
Getting started
Before you start
- Node 22 or newer
- Claude Code or Codex CLI
- a git checkout
- a GitHub account
- a GLAD install token (sign up)
01Install
GLAD ships from its own private registry. Point npm at it with your personal token, then install.
$ npm config set @abloh:registry https://npm.glad.abloh.dev/ $ npm config set //npm.glad.abloh.dev/:_authToken <your token> $ npm install -g @abloh/glad $ glad --version 0.1.0
Your token arrives when you sign up.
02Log in
Sign in with GitHub to start GLAD for this repo.
$ glad login Open https://github.com/login/device and enter the code 4F2C-9A1D Waiting for GitHub. service http://127.0.0.1:52144 as Kenneth <kenneth@example.com> human hum_3f9c2a71 arm it glad arm
The judge comes provisioned with it: no API keys to manage.
03Arm the repository
$ glad arm
armed /Users/kenneth/code/pay
wrote ~/.claude/settings.json, ~/.claude.json, ~/.claude/skills/glad/SKILL.md
service http://127.0.0.1:52144
glad open to see the graphArming is once per repository: GLAD goes into your own Claude Code settings, so every clone and worktree of this repository runs it, and no other folder does. GLAD keeps running through crashes and reboots, and glad disarm removes every trace.
glad arm --harness codex, then in codex: /hooks, press t to trust.
04Work as usual
Keep prompting exactly as you do now: GLAD reads your decisions from your own words and your agent logs its own, so there’s nothing extra for you to do.
Your agent makes hundreds of micro decisions it never mentions, and some carry real consequences: logging full card numbers to debug a webhook, soft-deleting users your privacy policy says are erased.
GLAD captures every one in real time and keeps your agent to what you decided. It asks you about two things only.
Your agent has a real reason to change your call
CONTRADICTION: users / account deletion you decided: erase data on account close I want: soft-delete, purge after 30 days because: Stripe disputes need the customer record keep yours, or change?
Your decision stands until you answer in your next prompt.
Two teammates clash: you both see it in real time
example/pay: auth / removed users: two people decided this differently. @Vladimir decided "revoke sessions on removal" (2026-10-04 10:12). @Logan decided "let sessions expire (24h)" (2026-10-04 10:14). Logan is being asked in their session; Vladimir's decision stands until they answer. [ Raise in Linear ]
When your own agent drifts for no good reason, GLAD steers it back without interrupting you.
05Trace any decision
Every decision from you, your teammates and every agent lands in one decision graph, with its whole context. “Why does it work like this?” always has an answer.
And it’s local: the graph lives on your machine, never on ours. We don’t store your code, your decisions or your prompts.
$ glad decisions users / account deletion answer soft-delete, purge after 30 days by you · 2026-09-12 16:40 status landed · main · v2.4.0 code src/users/close.ts#closeAccount:18-41 depends on billing / dispute window = 30 days · Logan replaced erase on close · you against Thomas's Claude Code agent · said: hard delete · coded: yes 1 in force · 1 replaced · 0 flags
No context lost, for anyone:
- People get every clash and every drift the moment it happens, right where they work.
glad openshows the whole graph. - Agents never start from zero: every prompt arrives with the decisions it touches.
Run glad doctor. See Something not working?
Get abloh IDE
abloh IDE is a Mac app for Apple Silicon and Intel: a multiplayer terminal IDE with GLAD built in.
Coming soon.
