Start here

Troubleshooting & FAQ

This page is organised by what you see, not by which part of Milly Lab caused it. Find your symptom, read the likely cause, apply the fix. It is for all Milly Lab users.

#Signing in and accounts

I never got the verification email. Check spam first, and mark the message as not-spam so future mail arrives normally. If it still has not arrived after a few minutes, request another from the sign-in page. If you signed up with Google you do not need to verify — Google has already done it.

My password reset link does not work. Reset links are single-use and expire. Request a fresh one. If you click an old link from an earlier email it will fail even though it looks identical.

I was signed out unexpectedly. Changing your password, or signing out anywhere, invalidates existing sessions everywhere — this is deliberate, so a stolen session cannot outlive a password change. Milly Lab also signs you out after a period of inactivity. Sign in again.

"Continue with Google" fails. Usually a blocked popup. Allow popups for Milly Lab and retry. If you originally signed up with an email and password, sign in that way instead.

#Messages and responses

The response stopped partway. Usually a network interruption. Resend the message. If it happens repeatedly on one model, switch models — a provider may be degraded.

Milly Lab checks before starting that a request can afford a complete answer, so a truncated response is not a budget problem. Being blocked before the response starts is.

The model gave me something confidently wrong. Expected behaviour of the technology, not a bug. Models generate plausible text and can invent facts, figures, citations and quotations. Verify anything that matters. If accuracy is critical, ask a second model the same question — Council Mode exists for exactly this.

Responses are slow. Check whether web search is on for the message. It adds a live search round trip and makes every message slower. Turn it off unless the answer needs current information.

Larger models are inherently slower. Long conversations are slower because the whole history is resent each turn.

The model forgot what we discussed earlier. Very long conversations exceed what the model can hold. Start a new conversation and paste in the key context, or switch to a large-context model such as gemini-3.6-flash-200k.

My uploaded file was ignored. Confirm the upload finished before you sent the message. Very large files may be truncated. Check the format is one the model can read — a scanned image of text is not the same as text.

#Payment-required errors

All arrive as HTTP 402 with a code. Full detail is in "Plans, Billing & Usage".

free_limit — Free tier, 10 messages per model per 4-hour window used up. Wait for the reset time shown, switch to the other free model, or subscribe.

free_model_locked — Free tier can only use gpt-5.4-nano and deepseek-v4-flash. Everything else needs a subscription. Balance cannot buy free-tier access to a paid model.

plan_required — The feature needs a paid plan of any kind.

feature_required — Your plan does not include this capability. File generation needs Pro or above; Swarm Mind needs Max or above.

insufficient_balance — That model's monthly allowance is spent and your balance cannot cover the next call. Switch to a model that still has allowance, or top up.

insufficient_for_request — This particular request is too large for what remains. Almost always a long conversation rather than a general shortage. Start a new conversation, remove large attachments, or shorten your prompt.

A model is greyed out. It is not in your plan. Plans are not cumulative, so a model available on a lower plan may not be on a higher one.

#Studios

My image or video job is stuck at "queued". Video generation genuinely takes minutes. The queue position is shown. Leave the page — the job continues server-side and appears in your history when it finishes.

If a job has made no progress for much longer than its queue position suggests, contact us through the Contacts page.

Generation failed with no clear reason. Usually the upstream provider rejected the prompt — often a content-policy refusal. Rephrase and retry. If every prompt on one model fails, that provider may be down; try a different model.

The result does not match my prompt. Be more specific about subject, style, composition and lighting. Use a reference image. Use the negative prompt for what you do not want.

My video controls changed when I switched model. Models support different resolutions, durations and audio. Switching clamps your selections to what the new model supports. Check them before generating.

Audio is greyed out. That video model does not support audio.

#Swarm Mind

A Project run is taking a long time. Project Mode runs many model calls across dependent subtasks. Live progress is shown per subtask. It is genuinely slower than chat.

A run cost far more than I expected. An orchestrator plus workers is many calls from one brief. If a workspace was connected, file contents were sent as input on top. Check Settings → Billing for the per-model breakdown.

I lost connection mid-run. The run continues server-side. Reopening the session replays events from where you left off, so nothing is lost.

"Too many concurrent runs". There is a cap on simultaneous runs per user. Wait for one to finish.

Council cards disagree. That is the point — independent answers to the same brief. Use the diff to compare and merge to combine the strongest parts.

#Connected workspaces

"Repository not found". Re-select the repository from the picker rather than relying on a previously typed name. For organisation repositories, the organisation may restrict third-party OAuth applications — an owner must approve Milly Lab.

Google says access is blocked, error 403 access_denied. The Google OAuth application is not yet verified, so only registered test accounts can connect.

The popup closed but nothing connected. The connection is bound to the browser that started it, by design. Use the same browser and profile throughout.

A file I expected was not read. For Drive, Milly Lab sees only files you opened with Milly Lab or that Milly Lab created. For GitHub, confirm the file is in the selected repository. Files over ~48 KB are truncated, not skipped.

Apply reported a conflict. Someone committed to that file after the run read it. Milly Lab left it alone rather than overwriting. Pull, re-run, apply again.

"An apply is already in progress". A previous Apply is still running. The claim clears itself after two minutes.

"Wait for the run to finish before applying changes". The run can still change what is staged. Let it finish.

Did Milly Lab change my repository without asking? No. Changes are staged and written only when you press Apply. If files appeared, an Apply was performed.

#Teams

My invitation was never received. Check spam. Invitations expire — ask the owner to resend. The account is not created until the invitation is accepted, so there is nothing to sign into beforehand.

I accepted an invitation but see nothing. Sign out and back in to refresh your membership.

The owner has left / we cannot manage the team. Ownership can be transferred to another member from the team settings page, so a team is never stranded.

#Files and documents

Document generation is refused. It needs Pro or above (feature_required).

My generated document will not download. Check whether your browser blocked the download. Try a different browser. If the link has expired, regenerate from the conversation.

#Frequently asked

Is my data used to train AI models? Milly Lab does not train models on your content. Your content is sent to third-party providers to produce responses; each has its own policy.

Can Milly Lab see my private repositories? Only if you connect GitHub. Read the Connected Workspaces Guide — the GitHub scope granted is broader than the single repository you select, and this is explained there in full.

Where is my data stored? On Milly Lab's hosting infrastructure — the Privacy Policy and Subprocessor List describe who processes what.

How do I delete my account? In Settings. Note that the current action deactivates rather than erases — the Privacy Policy explains how to request full removal.

How do I export my data? There is no self-service export yet — request one through the Contacts page.

Does unused allowance roll over? No.

What happens when my subscription ends? You drop to the free tier. Nothing is deleted.

Can I pay for a plan? Not yet — no payment gateway is connected. One with local banks is planned.

Is there an API? Yes. See the API Reference.

Is there a mobile app? Native iOS and Android apps are in development.

#Still stuck

Reach us through the Contacts page. Include what you did, what you expected, what happened, the model, and the time — it makes diagnosis much faster.