Skip to content

About

The second brain for you and your agents, on your own machine. One Docker install; your AI assistants connect over MCP, each under its own name. Guides for people and for agents.

Topics

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

Glassy Self-Host

Version Runs on Docker For AI agents License: MIT

Glassy's Today view: good morning, Sam, and two questions from an assistant called Juniper waiting for a decision

The second brain for you and your agents, running on your own machine.

Notes, documents, bookmarks, voice memos, calendars and boards on one calm page. It is the same page your AI assistants read and write, each under its own name. Your files stay on your disk.

This repository installs it. One container, about ten minutes.

Install · What you get · What you take on · For AI agents · Guides

Install

You need Docker on a 64-bit Intel or AMD machine (Windows, Mac or Linux), about 2 GB of disk, and a Glassy account. A free one is enough: sign up at glassy.fyi. Full requirements.

git clone https://github.com/0Reliance/glassy-selfhost.git
cd glassy-selfhost
./install.sh

On Windows, in PowerShell: .\install.ps1

The installer checks your machine, asks for your account email and pairing token, starts Glassy, and tells you how to sign in at http://localhost:3000.

The installer finishing: every check marked ok, Glassy running at localhost:3000, and the next steps

The full walk-through, including doing it by hand, is in docs/install.md.

What you get here

This is all of Glassy, the same software that runs on glassy.fyi. Running it yourself is where these three go furthest.

Your assistants work beside you

Give each assistant its own named key. It reads what you've saved, writes where you can see it, and asks you when a choice is yours. You can always tell who wrote what.

The notes view: cards written by Sam, and cards written by an assistant called Juniper, each marked with its author

A question from Juniper, 'Shall I book the Graça house for Lisbon?', with three choices as buttons, each saying what it leads to

Watch a 25-second film on Glassy Learn: who wrote that? Every note and document shows who wrote it, you or an assistant

AI runs where you choose

In your browser, on your own Ollama, or with your own key from a provider. Glassy provides no AI here and never bills you for any. With none set up, everything except AI still works.

With a model of your own, search finds things by what you meant as well as the words you typed.

A search for 'somewhere quiet to stay with no stairs' finds the note comparing three places to stay in Lisbon first, matched by meaning and by words

Your Obsidian vault, live

Keep writing in Obsidian. Glassy searches your vault beside everything else you save, and an edit in Glassy lands in the file. Tick a box in Glassy and it's ticked in Obsidian. How to link it.

A vault note, 'Garden plan', open in Glassy: the vault's Garden folder on the left, the note's checklist on the right with two items ticked

What you take on

  • You run the updates. Nothing changes until you choose. Update Glassy.
  • Backups are yours. Glassy saves a copy every night on the same disk. Keeping one somewhere else is up to you. Backups.
  • One person by default. More workspaces for you are free. More people need a seat pack.
  • It lives in Docker, using about 2 GB of disk plus your notes.

What leaves your machine

Diagram: notes, files, passwords and saved keys stay on your machine. Three things can leave: an account check at start, AI services you choose, and Cloud Sync if you turn it on.

When Glassy starts, it confirms your account with the Glassy cloud: your email and pairing token, no password, no notes. If that can't happen, Glassy starts anyway. After that, the only things that leave are the ones you turn on. There is no tracking. Security and privacy.

For AI agents

Handing this to an assistant? Give it AGENTS.md. It routes an agent to one of four sets of steps: install Glassy, set it up, work in it, or fix it. Each says what to ask you first, when to stop, and how to show the job is done.

Diagram: each assistant holds its own named key and connects to Glassy's agent endpoint; its work lands in your workspace, the archive, or its own space

Connecting an assistant yourself takes a minute: Agents in your Glassy.

Guides

I want to… Read
Install Glassy docs/install.md
Find my way around in the first ten minutes docs/first-steps.md
Choose where AI runs, connect an assistant, change settings docs/configure.md
Understand AI and search here docs/ai.md
Work with assistants docs/agents.md
Link my Obsidian vault docs/obsidian.md
Use Glassy on my phone or another computer docs/reach-it-from-other-devices.md
Keep a second copy in sync docs/cloud-sync.md
Run two Glassys of my own, with no Glassy server between them PAIRING.md
Back up and restore docs/backups.md
Update, or see what changed docs/upgrade.md, docs/whats-new.md
Sort out my account or a forgotten password docs/accounts.md
Fix a problem ./doctor.sh, then docs/troubleshooting.md
Know what is private docs/security.md

Learn Glassy itself, step by step, at learn.glassy.fyi. The full product documentation is at docs.glassy.fyi.

Help

Run ./doctor.sh first. Then open an issue for a problem with installing or running Glassy yourself. Assistants file theirs with the "Agent report" form.

The files in this repository are produced from Glassy's main source on every release, so changes made here directly are replaced. Contributing says how to suggest one.

License

The Docker Compose files, Caddy configuration, scripts and documentation in this repository are licensed under the MIT License.

The Glassy application image (ghcr.io/0reliance/glassy-dash) pulled by these compose files is separately licensed under the Business Source License 1.1 (BSL 1.1). It is available for personal, non-commercial, self-hosted use and will transition to AGPL-3.0 on 2027-07-13.

For commercial licensing: [email protected]

About

The second brain for you and your agents, on your own machine. One Docker install; your AI assistants connect over MCP, each under its own name. Guides for people and for agents.

Topics

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages