I built a 3D office where every Claude Code session I run is a little person at a desk
retained text
I use Claude Code with a personal account and my work account, and kept losing track of which terminal was doing what, which one was stuck waiting for me, and how much of each account's window I had used. So I built clalt, a small Go CLI that keeps each account in its own sessions, and gave it an "office": a 3D page in the browser where every running session is a character. What the office shows, all from your real sessions: * one character per session, sitting at a desk in its account's colour * what it is doing right now: leaning in to read, typing fast on an edit, leaning back while a command runs, a hand to the forehead when it fails, arms up when it passes * when a session needs you, it stands up and waves until you answer * subagents get their own desks; when one finishes it walks over to its main agent and reports back * a meeting room and a review room upstairs for planning and review agents, breaks at the cafe when idle (or hide them, your choice) * per-account 5h and 7d usage bars, a Today panel and a history chart * PR reviews your agents are working on The CLI side: `clalt login <name>` signs an account in with Claude's own sign-in, `clalt run <name>` starts claude as that account (any number in parallel), `clalt usage` and `clalt dashboard` show usage per account. How I built it with Claude Code: I built clalt with Claude Code over about two months (Go for the CLI and the local server, three.js for the office). I decided what to build and reviewed everything; Claude Code wrote most of the code. What worked for me: * plan first: for each feature Claude wrote a short plan and asked me the design questions before touching code * tests first: every bug fix and feature started with a failing test, which kept a large codebase (about 2,700 Go tests and 870 JS tests) from regressing * one change per branch and pull request (96 merged so far), with Claude reviewing its own diffs before I merged * parallel subagents for independent pieces, and Claude driving a real browser (Playwright) on fake data to check the 3D office, which caught bugs the unit tests missed, like an agent walking over to brief an empty chair * before going public, Claude checked the code and docs against Anthropic's terms and scrubbed personal data from the history Free to try: it is free and open source (MIT), no paid tier. Install it from the releases page or with `go install`, then `clalt login <name>` and `clalt office`. It runs locally only: the office listens on 127.0.0.1 behind a random token, reads your transcripts on your machine and sends nothing anywhere. It runs the official claude binary and signs in through claude's own login, so it never sees a token. It is not a way around usage limits: every account has to be your own and stays under its own terms. macOS and Linux, Windows best effort. Unofficial, not affiliated with Anthropic. Repo: [https://github.com/dhitalkamal/clalt](https://github.com/dhitalkamal/clalt) (MIT) Install: download a binary from the releases page, or go install github.com/dhitalkamal/clalt@latest Happy to hear what you would want an office like this to show. Edit: Add 4 more template option of layout.
document fetched 2026-11-04 00:39 · canonical https://reddit.com/r/ClaudeCode/comments/1xmym86/i_built_a_3d_office_where_every_claude_code