whipcode
How it works

One hub for every coding agent on your machines.

Whipcode is a self-hostable hub plus a small client that runs on each of your machines. The client connects out to the hub over a secure WebSocket, so your machines need no public IP and no open port.

From there, every agent conversation streams into one web workspace: text, tool calls, subagents, diffs and the questions that are waiting on you. Here is what happens, step by step.

Before you connect: have a Mac or Linux machine with your repository, Node.js 22 or later, and the coding CLI you want to use installed and authenticated. Keep that machine online and awake. Managed access and repository access are by invitation during the private beta; model usage follows your providers' plans and limits.

01 / The shape of it

Clients connect out to one hub.

You run a lightweight client on each computer that holds your repos and CLI logins: a laptop, a desktop, a Linux server. Each client dials out to your hub. The hub serves the workspace you open in a browser or on your phone.

Because the connection is outbound, your agent machines stay behind their router. No inbound port, no exposed IP. You choose where the hub lives: the managed beta, or your own server.

Client on each machine → your hub → you

02 / Follow the work

Every conversation streams live.

Open the workspace and each agent shows its work as it happens: the text it writes, the tools it calls, the subagents it spawns, the diffs it proposes. You read the same stream whether the agent runs on the Mac next to you or the server in another room.

The engines you already use plug in as they are: Claude Code, Codex, Grok, Gemini, Kimi and Muse. Pick one per conversation, or let Auto route a prompt using the available models and your engine preferences.

Live output · Tool calls · Subagent activity · Questions · Per-conversation model switching
03 / Answer what needs you

Questions arrive as answerable cards.

An agent waiting for an answer is an agent doing nothing. When any conversation needs a decision, Whipcode surfaces it as a card you can answer in place, across every connected machine, from one inbox.

You can attach a screenshot or a file to give the agent the context it needs, right in the conversation. Read the context, make the call, and the work keeps moving.

04 / Delegate the follow-through

Give a longer job a mission.

Some work takes more than one turn. A mission lets an agent keep going toward a goal without waiting for your next prompt. A separate AI judge checks the result against that goal and sends it back for another pass when something is missing.

Missions run within configured limits. They can hit a limit, get blocked, or reach a point that needs your decision, and you can inspect progress, pause or stop at any time. The judge adds a check; it does not guarantee correct code or replace your review.

Goal-oriented · AI-judged · Watched for stalls · Pause and stop anytime
05 / Step away, stay in the loop

Reach the same conversations from your phone.

Start work at your desk and check in from a mobile browser built for a smaller screen. Answer the question that is holding things up without reopening your laptop. Add the workspace to your home screen if you like.

The agents keep running on the connected machine you chose. That machine has to stay online and awake; closing the browser does not move the job to another computer or to the cloud.

06 / Keep your terminal

The CLI is still right there.

Whipcode sits beside your terminal, not on top of it. Claude conversations use the CLI's own session files, so you can pick one up on the machine with claude --resume <session-id>. Classic tmux sessions, one isolated socket per project, are reachable from the hub when a job calls for a shell.

Session continuity depends on the engine, so the Claude resume behavior is not a promise for every provider.

07 / Where it runs

Managed beta, or self-host it yourself.

The managed hub is in private beta. If you prefer to run your own, the core is MIT licensed, with repository access by invitation during the beta. A self-host comes down to a few steps on your server:

git clone the repository, then npm install && cp .env.sample .env && npm run build && npm run start:web. On each machine, start the client pointed at your hub over wss://, then open the hub's root address (/) on desktop or phone. Whipcode Mobile uses the same entry; older /cvt and /cvtm links remain for compatibility.

Logins, terminal input, file transfers and user changes are written to an audit log, with a dangerous-command blocklist and rate limiting on login and machine connections. Engines use the CLI authentication configured on each machine, subject to your providers' plans and limits.

They code. You lead.

See the use cases Whipcode is built for, or request access to the private beta.

Get early access Explore use cases → Private beta · Engine usage billed by your providers.