Claude Code isn’t signed in
The title names the chat’s agent: Claude Code isn’t signed in or Codex isn’t signed in. The agent’s sign-in ran out, or it was never signed in on this computer.- Click Sign in to Claude or Sign in to Codex and finish signing in. After a turn failed, the button says Reconnect Claude or Reconnect Codex.
- If the card shows Open settings instead, sign in under Provider accounts at the top of Settings > Agents.
- If you have another account of that agent signed in, click Switch to followed by its name and retry to send your message there.
Account usage limit reached
Your Claude or ChatGPT account reached a limit its plan sets, such as its 5-hour or weekly limit. These limits come from Anthropic and OpenAI, not from Wonder, and count everything you do with the account. When Wonder knows the time, the card starts with it, for example Resets 3:50 PM.- Switch to an account and retry: sends your message again on another Claude account you’ve signed in, and shows how much of its limit is left. Switching is offered in Claude chats only.
- Continue with a summary: starts a new chat from a summary of this one.
- View usage: opens Settings > Usage, with each limit, how much you’ve used and when it resets.
You’ve used this week’s AI layers
On the Free plan, your agents can make 3,000 layers on the canvas a week. At 80% the message box shows You’ve used followed by the share, and once they’re all used, You’ve used this week’s AI layers, with the time they reset. The chat itself doesn’t stop: you can keep chatting and working on code, but the agent can’t add layers to the canvas until the reset. When it tries, it passes on Wonder’s message that the limit is reached. To keep designing, click View Pro and upgrade the workspace to Pro, which has no weekly limit. See 3,000 AI-generated layers a week.This conversation is too long
The conversation filled the agent’s context window, so it had no room left for your message. Wonder often fixes this by itself: the card says This conversation got too long and Wonder is compacting it to make room, which replaces older messages with a summary. When it’s done, it sends your message again. If the turn had already done some of its work, Wonder doesn’t resend it: ask the agent to continue. When the card says This conversation is too long:- Click Compact and retry to compact the conversation and send your message again.
- Click New chat to continue in a new chat, which starts from a summary of this one.
This chat now runs contained
The chat worked on a pull request from a fork, or took in text from a chat that did. Anyone on GitHub can write a fork’s code, so from then on the chat runs contained: it asks before more of what it does, it can’t use Full access and its approvals aren’t remembered. The chat keeps working; nothing is broken. A chat stays contained. For work that doesn’t involve the fork, start a new chat in a worktree that doesn’t hold its pull request. See Chats that worked on a fork’s pull request in Permission modes and approvals.Disk almost full
The sidebar shows Disk almost full when this computer has less than 15 GB free. New worktrees and repositories may fail until you make room.- Click Free up space to delete your archived worktrees. It only shows when you have some.
- Otherwise, free up space on this computer yourself, then try again.
Claude Code isn’t installed on this computer
Wonder installs and updates Claude Code and Codex for you. When the chat’s agent isn’t on this computer, the title names the agent, for example Codex isn’t installed on this computer. Click Install to install it and send your message. If the install fails, the message box shows why, with Retry install:- could not be downloaded. Check your internet connection, and that no proxy or firewall blocks the download, then try again.
- There isn’t enough free disk space to set up the agent. Free up space, then try again.
- Something on this computer, such as security software, stopped the agent from starting. Allow it to run, then try again.
- Memory on this computer is low, so the agent couldn’t finish its startup check. Close apps you aren’t using, then try again.
- could not run on this computer. Check for a Wonder update. Click Update Wonder. See Update the desktop app.
The chat couldn’t connect
These cards mean the agent couldn’t reach its service or start the chat:- Claude Code couldn’t connect, Claude Code couldn’t reach its service, Claude Code’s service didn’t answer in time or Claude Code lost its connection to its service (or the same with Codex). Check your internet connection. A firewall, VPN or proxy may be blocking the agent. Then click Try again, or send a message. If the turn had already done some work, ask the agent to continue instead.
- This chat couldn’t start, Chat connection timed out, Claude Code stopped responding or Canvas tools didn’t connect. Click Retry connection. If it fails again, click Restart Wonder.
- The turn couldn’t finish or The chat stopped. Send your message again. If it keeps happening, click Restart Wonder.
Other cards
- This model isn’t available. The model doesn’t exist or this account can’t use it. Pick another model, then send your message again. See Choose models and reviewers.
- The model’s safeguards flagged this message. This sometimes happens with safe, normal requests. Pick another model for this chat, then send again.
- This turn reached its step limit. The agent stopped after 500 steps in one turn, so a loop can’t run on. Ask it to continue if it was making progress.
- The chat’s agent can’t read its settings. There’s a mistake in the agent’s own settings file. Click the button to open it, fix it, then send again.
If something goes wrong
If a chat keeps stopping and none of the above helps, export Wonder’s diagnostic logs and send them to us:- Choose Help > Export diagnostic logs.
- Pick where to save the file and click Save logs. It holds technical diagnostics, without your chats, repository files or credentials.
- Email it to team@wonder.so with what you were doing. See Support for what else to include.