VCode home

Claude Code hit its limit. Keep going in Codex, or for free.

On Claude's Pro plan, Claude Code gets a 5-hour usage window and a weekly one. When one runs out in the middle of your work, you can wait for the reset, pay for extra usage, or hand the work to another agent. VCode does the last one from a button under the message box. Pick Codex if you pay for ChatGPT. Pick OpenCode to keep going for free. The new agent gets a written summary of the work so far and carries on in the same tab.

Your options when the limit hits

OptionWhat it costsWhat happens
WaitNothingClaude Code carries on after the reset. In VCode the tab pauses and picks the work up by itself ten minutes after the reset.
Extra usagePay as you goAnthropic sells usage past your plan's window. Claude carries on where it stopped.
Hand it to CodexA ChatGPT planCodex reads a summary of the work and carries on in the same tab.
Hand it to OpenCodeFreeThe same, on a free model. Free models are weaker than Claude.

Waiting is often the right call. Forty minutes left on the 5-hour window is a coffee break. Two days left on the weekly window is not. The meters, further down, tell you which one you have.

Hand the tab to another agent

These steps are the same for every agent. The sections after this one say what each agent needs first.

  1. When the limit hits, Claude has already stopped. If it is still working, stop it first: a tab cannot change agent mid-task.
  2. Tap the agent button under the message box. It shows Claude.
  3. The menu opens, headed Agent for this tab. Claude's row reads running this tab now. Every other agent's row reads hand this tab over; it reads the transcript as a brief. Tap the one you want.
    The agent menu open over the message box. The header reads AGENT FOR THIS TAB. Claude's row says running this tab now. The other rows say hand this tab over; it reads the transcript as a brief.
    The agent menu. Tap the agent you want.
  4. A Sys row appears in the conversation. Tap it to read the note, for example handed off from claude to codex; codex gets a brief of 18 items with your next message.
    A Sys note in the conversation reading: handed off from claude to codex; codex gets a brief of 2 items with your next message.
    The note says how much of the conversation the summary holds.
  5. Type what you want next and send it. The summary goes in front of your message, once.

What the new agent gets: your messages, Claude's replies, and one line for each command Claude ran and each file it changed. The output of those commands is left out, so the new agent runs again whatever it needs. The summary keeps the newest 24,000 characters. On a long conversation the start drops off.

The new agent starts its own session. It cannot open Claude's. Claude's replies stay in the tab with Claude's mark on them, so you can see who did what.

Codex: if you pay for ChatGPT

Codex is OpenAI's coding agent. It comes with ChatGPT's paid plans. It has its own usage windows, separate from Claude's.

  1. On the workstation that runs VCode, install Codex:
    curl -fsSL https://chatgpt.com/codex/install.sh | sh
    Use this installer, not npm install -g. VCode can update Codex every hour, but not a copy npm installed with sudo: that update needs root.
  2. Run codex once in a terminal and sign in with your ChatGPT account.
  3. In the agent menu, tap Codex.

OpenCode: free, no account

OpenCode is an open-source coding agent. Its own service, Zen, has free models that need no account and no key. A new OpenCode starts on one of them, Big Pickle.

  1. On the workstation that runs VCode, install OpenCode:
    curl -fsSL https://opencode.ai/install | bash
  2. In the agent menu, tap OpenCode. You do not need to restart anything.
  3. Send your message. The tab runs on Big Pickle. To pick another free model, tap the model button next to the agent button. The free ones are opencode/big-pickle and the names that end in -free.

The free models change as providers come and go, and they are weaker than Claude. Give them small, clear steps, and read what they change before you build on it.

See the limit coming

A row of meters sits at the top of the screen: one card for Claude, one for Codex. Each card shows the 5-hour window (5h) and the weekly window (W): how much you have used, and how long until it resets, such as 42m or 2d 4h.

The meter row: a Claude card and a Codex card, each with a 5h row and a W row, a percentage, a bar and the time until the window resets.
Claude and Codex side by side, each with its 5-hour and weekly windows.

OpenCode has no card.

When the window runs out in the middle of a task, the tab does not fail. A line appears at the end of the conversation saying when the tab will carry on, for example Claude usage limit reached. Your queue will pick up at 14:20. Under it, it names the window and counts the messages waiting, and the time left counts down beside the Retry now button.

The usage limit line: Claude usage limit reached. Your queue will pick up at 17:53. Under it: 5-hour window ยท 1 message waiting. Beside it: 1h 12m and a Retry now button.
The tab is paused, and it says when it will carry on.

Leave it, and the tab picks the work up by itself ten minutes after the reset. Send a message or hand the tab over, and the wait is cancelled.

On your phone

Everything above works on a phone at app.fezu.dev. The meters are at the top, the message box at the bottom, and the agent button sits under the message box.

The whole phone screen: the meters and the tabs at the top, the conversation with the usage limit line near the bottom, and the message box and the agent button at the bottom.
A Claude tab paused on its limit, on a phone.

Keep a Claude tab and a Codex tab open side by side, and swipe the conversation sideways to move between them. From the last tab, one more swipe opens a page for starting a new tab.

A swipe in progress on a phone: the conversation pushed aside with the next tab's page coming in behind it.
Half way through a swipe to the next tab.

Each tab keeps its half-written message. Type in one tab, swipe away, come back, and the text is still there.

Set up VCode

You need a Linux PC, a Windows PC or a Mac with Claude Code on it.

  1. Open a terminal on that computer (PowerShell on Windows, Terminal on a Mac) and run the install command. On Linux or a Mac:
    curl -fsSL https://fezu.dev/install | sh
    On Windows:
    irm https://fezu.dev/install.ps1 | iex
  2. When it finishes, VCode opens. Log in or create an account with your email; we send a code to it. A new account gets a 30-day free trial.
  3. Scan the QR code VCode shows with your phone's camera. Your phone opens VCode on the same account.
  4. Start a tab on Claude and work as usual. When the limit hits, hand the tab over.

FAQ

Does the new agent see what Claude did?

It gets a written summary in front of your next message: your messages, Claude's replies, and one line for each command Claude ran and each file it changed. The output of those commands is left out. The summary keeps the newest 24,000 characters of the conversation.

Do I have to pay for another AI plan?

No. OpenCode's free models need no account. Codex needs a ChatGPT plan.

What does VCode cost?

VCode is free for 30 days, then $10 a month or $96 a year. Prices in pounds and euros are on the pricing page. You can also pay once for Lifetime. We sell a limited number of Lifetime plans. The agents' own plans are separate: VCode runs the agents you have and sells no AI usage of its own.

Does my workstation have to stay on?

Yes. The agents run on it, and your phone shows that workstation's tabs. Switch it off and the tabs stop.

Does it work on iPhone? On a Mac?

iPhone: yes, in Safari at app.fezu.dev. Mac: yes. The workstation that runs the agents can be a Linux PC, a Windows PC or a Mac, Apple silicon or Intel.