Open source · MIT · self-hosted

Browser Tabs for Claude Code

Paddock is a self-hosted home for your Claude Code agents — source included, MIT, running on hardware you own.

A dozen conversations at once, each in its own project, none of them lost when you close the lid. Chats persist, resume, fork and rewind.

View on GitHub Developer docs Blog posts Watch the video

Deploy anywhere

$ npx @edspencer/paddock -o

Install Node, then open Paddock on your own Claude Code history. Nothing is written into your directories. How Discover works

How it works

Chats that outlive the terminal.
Work that starts without you.

Start from the history you already have

Paddock opens on Discover: it reads the Claude Code history already on your machine, lists the directories you have been working in, and turns the ones you tick into projects — conversations and all. Nothing to clone, and nothing written into your directories.

Every chat belongs to a project

A project is just a directory — a codebase, a research notebook, a home-ops runbook — and Paddock runs Claude Code in it. Projects sit in the rail, in whatever groups you give them, each carrying its own unread count. Pick one and you get its chats, and only its chats.

Paddock’s project rail with six projects under two headings, Homelab and Side Projects, several showing unread counts; Trail Atlas is selected and the column beside it lists that project’s five chats

Point a project at a git repo

A project can be a plain directory or a clone of a repo you own. Either way the agent works on a real checkout, and the Changes tab shows you the actual diff — stage what you want, leave the rest, commit and push from the browser.

A Paddock project’s Changes tab: branch main with 3 uncommitted files, a changed-files list with all three ticked, and an inline red/green diff of src/render.ts

Chats open other chats — in other projects

A conversation is not a dead end. Claude can open a fresh chat and hand it a kickoff prompt, in this project or a different one, and each new chat records the parent it came from. Work that belongs somewhere else goes and happens there.

Two Create chat tool blocks in a Lumen CLI conversation, each badged TRAIL-ATLAS, reading “Created chat … in trail-atlas” with the kickoff prompt each new chat was handed

It sends things back

Not just text. The agent can hand you a diagram, a document, an image or a chart and it renders in the conversation where you can read it — rather than a filename you have to go and open somewhere else.

Two files sent by the assistant and rendered inline in a Paddock chat: a Mermaid flowchart of a colour pipeline, and a Markdown document titled “How a seed colour becomes a theme”

Watch the work, not just the answer

File writes, commands and nested sub-agents render as live cards as they run — current step, step counts, durations. Here one chat runs two research sub-agents in parallel and writes a haiku while they work.

Fork it, or rewind it

Sessions live on the server, so they survive reloads, reconnects and devices. Hover any message to see how old it is and how full the context window was at that point — then branch off it. Name the branch for the question it goes off to answer, and it lands under its parent with the transcript ending where you split it. Or roll this one back instead.

See it work

Paddock in 90 seconds

A dozen conversations at once, running on a box in the corner of the room — forked, rewound, scheduled, and picked back up on a phone.

Get started

Try it on the history you already have

Paddock opens on Discover: it reads your existing Claude Code history, lists the directories you have been working in, and turns the ones you tick into projects — with their conversations intact. Nothing to clone, and nothing written into your directories.

Deploy anywhere

$ npx @edspencer/paddock -o

Install Node, then open Paddock on your own Claude Code history. Nothing is written into your directories. How Discover works

View on GitHub Developer docs Blog posts Watch the video