open source · self-hosted

Your daily
mug of AI.

DayMug is an open-source workspace for Claude and Codex that you host yourself — shared by your whole team on laptop, phone or tablet, and right inside your chat apps. Together. In sync. Yours to keep. Pour one and pass it around.

Set up
in 5 minutes
Works on
any browser
For
the whole team
Source
open on GitHub
01 live agent
02 served live
your daily mug
§ 01 — what's inside

A shared room for your AI, not a single seat.

AI is amazing when you're alone with it. DayMug makes it amazing when you're not. Bring everyone into the same conversations, pick up where someone else left off, keep your work close at hand.

01 / 06 // for the team

Skills brewed once, shared by all.

Give every person their own AI — with its own name, its own job, its own rules about what it can touch. Set up a skill once and the whole team can reach for it — no copy-pasting prompts between teammates.

ready-made roles shared skills simple permissions
02 / 06 // live

Watch it, pause it, pass it along.

See the answer arrive as it's being thought of — including the thinking. Two people can watch the same conversation at once. Stop it mid-sentence, slip in a correction without starting over, or hand it to a colleague and pick up later from another device.

live answers watch together cut in mid-thought
03 / 06 // yours to keep

Your work stays with you.

Conversations, files, history — they all live on a computer you choose. No tracking, no middleman, no surprise bills. Updates land with one click, and you can go back if something feels off.

no tracking your data undo updates
04 / 06 // anywhere

It's just a website.

Open it in any browser — laptop, phone, tablet, the cafe screen. Nothing to download. The conversation you started on the couch is still right there when you sit down at your desk.

any browser mobile-ready no app needed
05 / 06 // where you already are

It answers in your chat app, too.

Put an agent in Slack, 飞书, Telegram or 微信 and talk to it there. Each thread keeps its own memory, replies stream into the message as it works, and the same conversation is waiting in the browser when you want the bigger screen.

slack & 飞书 telegram & 微信 thread = conversation
06 / 06 // bring your own

Your AI, your accounts, your rules.

Claude and Codex both work, side by side, picked per conversation. Share one subscription across the team and the next person waits in a visible queue, or give everyone their own. Agents can be fenced into a single folder so nobody wanders into someone else's files.

claude & codex shared or separate fenced folders
§ 02 — what it looks like

A workspace,
not a conversation window.

Your conversation on one side, the files being worked on right beside it. Switch between your agents with a click. See what's happening in real time — no need to ask "did you change that yet?"

  • Switch between teammates instantly — each keeps their own conversations and history
  • Watch the answer come together as it's being thought of
  • Browse and preview files in the same window, no jumping around
  • Hand off mid-conversation — they pick up where you stopped
§ 03 — usage

Three sips and you're brewing.

From signing up to your first conversation in about a minute. Nothing to read, nothing to memorise — you'll figure it out as you go.

step / i.

Pick an agent

Choose an agent for the job. Define its role, and decide exactly what it's allowed to touch.

— an agent name: "Writer" role: edit drafts perms: read/write
step / ii.

Pick a working folder

Point each conversation at a folder. Pick a different one next time — every conversation stays in its own lane, with no spill-over between projects.

— this conversation folder: ~/welcome-emails — next conversation folder: ~/release-notes
step / iii.

Start the conversation

Type what you want and watch the answer come together. Hand it off to a teammate, or pick it up later from another device — the conversation is waiting where you left it.

you soften welcome email writer reading welcome.md writer saved — try it?
§ 04 — run your own

Open source.
One binary. Your machine.

The whole thing is on GitHub. Download a release or build it yourself — a Go server with the web app embedded and SQLite inside. No Docker, no external database, no cloud account.

  • Claude (Agent SDK) and Codex (app-server), side by side
  • Any Anthropic- or OpenAI-compatible endpoint, too
  • Slack, 飞书, Telegram and 微信 — no public URL needed
  • SSO, per-user sandbox, cost guardrails, one-click upgrades
  • No telemetry — your conversations never leave your machine
quickstart  — linux x86_64 · apple silicon
# 1. prerequisites: Node.js 18+ and the Claude Agent SDK
$ npm i -g @anthropic-ai/claude-agent-sdk

# 2. install into ~/.daymug (runs as your user, not root)
$ curl -fsSL https://daymug.com/install.sh | bash

# 3. put your username under admin.bootstrap_usernames
$ nano ~/.daymug/config.yaml

# 4. start it and open http://127.0.0.1:8090
$ systemctl --user start daymug.service
§ 05 — questions

Frequently brewed.

Short answers to the things people usually wonder about. If yours isn't here, drop us a line.

Q.01 Why use this instead of just talking to AI on my own?

AI on its own forgets. DayMug remembers — your conversations, your files, what your team has been working on. It's the difference between asking a stranger and asking a colleague who knows the room.

And because it's a website, you don't need to learn anything new. Open the link, talk to your agent, hand the conversation to a teammate when you're done.

Q.02 Where does my data live?

On a computer you choose — conversations, files and history sit in a SQLite database and folders on that machine. There is no telemetry or analytics. DayMug only talks to the AI providers and chat apps you connect, plus the release server when an admin checks for updates.

Q.03 Do I need to pay for the AI separately?

Yes — DayMug works with whatever AI plan you're already on. Claude and Codex both plug in, and you can run them side by side, choosing per conversation. We don't sell credits or stand between you and your agent; DayMug just adds the friendly room around it.

Q.04 Can a team share one AI account?

Yes. If one teammate is using it, the next one waits in a polite line — and they can see how soon they're up. Or give each person their own account if you want stricter separation. Whatever fits.

Q.05 Can I talk to it from Slack or 微信 instead?

Yes. Point an agent at Slack, 飞书, Telegram or 微信 and it answers there — each thread is its own conversation with its own memory, and the reply fills in as it works. Nothing is exposed to the internet to make this happen: DayMug dials out to the chat service, so it works from a laptop behind a router.

The same conversations are in the web app, so you can start something in a thread and finish it on the big screen.

Q.06 Does it really work on my phone?

Yes — open it in your phone's browser and you'll get a layout made for the smaller screen. Catch up from the train, the couch, the queue at a cafe. The conversation is the same one you started on your laptop, just sized for your hand.

Q.07 Is it really open source? Can I change it?

Yes — the full source, backend and web app, is on GitHub. Build it yourself, read how it works, and send issues or pull requests. The docs cover installing, configuring and running your own instance.

— still curious? open an issue on GitHub or write to [email protected]

Ready to pour one?

Five minutes to set up. Zero minutes to learn. Open source, on your own machine — for the whole team.