Troubleshooting

TL;DR

• Symptom-led — jump to the heading that matches what you’re seeing.
• Try the cheapest fix first (logs → re-prompt → restart → reset).
• If three attempts don’t unstick it, start a new chat — context is usually poisoned by then.

​

Jump to

• Fix a preview that never loads
• Break the agent out of an error loop
• Recover from a regression
• Fix connector auth errors
• Unstick “I cannot do that”
• Fix external build timeouts (Vercel / Netlify / EAS)
• Wait out EAS queue delays
• Cut runaway credit spend
• Bring a published URL back online
• When all else fails: start over

If your specific problem isn’t here, the FAQ and Support policy cover broader categories.

​

Fix a preview that never loads

Most common cause: the dev server crashed mid-build. Recovery, in order of cost:

1. Open the Logs drawer and look for a stack trace. It’s almost always something simple — missing dependency, syntax error in the most recent file write.
2. Ask the agent directly: “the preview is blank, check the logs and fix it.” The agent reads the logs and patches it.
3. If the sandbox itself is wedged: Project menu → Sandbox → Restart. Your files are preserved; the runtime is rebuilt fresh.
4. If restart didn’t fix it: Sandbox → Reset. Destroys the running sandbox and creates a new one from your files. Slower than restart, but recovers from rarer failure modes.

Still stuck? → File a ticket via Support policy.

​

Break the agent out of an error loop

Three turns trying to fix the same TypeScript error is your signal to stop and re-prompt. The loop has poisoned its own context. What to do:

• Be more specific. Paste the exact error text. Name the file and line.
• Switch to Plan Mode and ask for a different approach: “Plan how we’d fix this without using the type assertion the last two turns kept trying.”
• Roll back the failed turns. Bad attempts in the chat history make the agent more likely to repeat them. “Revert the last 3 turns to and try again with a clean slate.”

Still stuck? → Start a new chat.

​

Recover from a regression

Roll back rather than patch. Almost always cleaner.

“Revert the last change to pages/Settings.tsx.”

Beats “fix the bug you just introduced” every time. If you’d already pushed to GitHub before the broken turn, that’s your safety net — branch off the green commit. Still stuck? → Iterating → Sync to GitHub when you’re in a green state.

​

Fix connector auth errors

Tokens expire and revocations happen.

• OAuth connectors (Slack, HubSpot, Notion) — open the Connectors drawer and re-authorize. A scope change on the upstream side can break previously-working calls; re-connect with the new scope.
• API-key connectors — re-paste the key in the drawer. Especially common after key rotation.
• Webhook-based connectors (Stripe, GitHub) — verify the webhook URL is still pointing at your project and the secret matches.

For MCP / Chat connectors specifically (when the agent says it can’t read your Linear / Figma / Notion mid-build), the auth lives in a separate panel — see MCP connectors → Where credentials live. Still stuck? → Auth & secrets.

​

Unstick “I cannot do that”

Usually the tool isn’t available in the current mode or project type.

• Plan Mode is read-only. Switch out of Plan Mode if you want a code change.
• Some tools are project-type-specific. Mobile-only tools won’t be available in a web project, etc.
• A tool may be disabled in Project settings → AI → Restrict tool use.

If none of those apply, provide an alternative path: “if you can’t do X, do Y instead.”

​

Fix external build timeouts (Vercel / Netlify / EAS)

Vibely’s preview works but the external build fails. The failure is usually unrelated to Vibely.

• Node version mismatch. Your hosting platform may default to an older Node than the project needs.
• Missing env vars. Local env vars aren’t carried to your host — set them at the host level.
• Missing build secrets. EAS, in particular, needs signing credentials registered before a production build will succeed.

Read the deploy log; the answer is almost always in there. If you’re truly stuck, paste the log into a Vibely prompt — the agent will diagnose it. Still stuck? → Mobile App → Ship for EAS specifics; Web App → Publish for web hosts.

​

Wait out EAS queue delays

The free tier has shared build queues. During peak hours, queue times can stretch.

• Check the EAS dashboard — if your build is “queued”, that’s a wait, not a failure.
• Vibely doesn’t have a knob for this; it’s an EAS infrastructure question.
• For predictable build times, upgrade to a paid EAS tier (priority queue).

​

Cut runaway credit spend

Usually one of three things:

• A runaway prompt loop the agent didn’t catch — kill the turn and reset the conversation.
• A leaked dev server keeping the sandbox warm 24/7 — restart the sandbox to recycle it.
• Smart routing keeps picking an expensive model — pin a smaller one from the model picker in the chat composer if you’re doing routine work.

The Usage drawer in each project breaks it down per turn so you can spot the source. Still stuck? → Plans & credits.

​

Bring a published URL back online

• DNS issue? Custom domains can have transient propagation issues. The Domains panel shows DNS health.
• TLS cert expired? Vibely auto-renews via Let’s Encrypt; if a renewal failed, the panel shows the error. Re-verify is usually the fix.
• Restricted access mode? If you switched to Workspace or Password access, the public URL now requires auth. Check Publish → Access.

Still stuck? → Web App → Publish.

​

When all else fails: start over

Start a new chat. Paste in the lessons from this one. It’s often the cheapest path forward — see Iterating → Know when to start over. If that doesn’t help, file a support ticket via the in-app chat or contact@chatmana.in. Response times by plan are in Support policy.

​

Next