Troubleshooting and FAQ

Symptom	Cause	Fix
”Your access is pending approval by an administrator.”	Your email is not approved for sign-in.	Ask your administrator. A super admin approves it in Platform access. Your attempt is already listed there as a pending request.
”Your organization (slug) requires single sign-on. Use “Continue with SSO” above.”	Your organization enforces SSO.	Use the Sign in with SSO button on the sign-in card.
No SSO button appears	The button only shows for emails that are already members of, or have a pending invite to, an organization with SSO.	Ask your administrator to invite you, or sign in with an email code if SSO is not enforced.
”Okta denied access. Ask your admin to assign you to the CorePillar app in Okta.”	You are not assigned to the app in Okta.	Your Okta administrator assigns you.
”SSO session expired. Try signing in again.”	The sign-in took too long or cookies were blocked.	Start again; allow cookies for the site.
”Enter the 6-digit code we emailed you.”	The code was not six digits.	Re-enter it. Request a new one if needed.
Code never arrives	Mail filtering, or a typo in the address.	Check spam, click Use a different email, and try again.
”Invite not found or expired”	Invite links expire after 7 days, or it was revoked.	Ask for a new invite.
”This invite was sent to a different email address”	You are signed in with another email.	Sign out and sign in with the invited address.
You land on an onboarding screen after signing in	Your email is approved but you are not a member of any organization.	Ask your administrator to add you. Do not create an organization unless you mean to.

The chat

Symptom	Cause	Fix
”No models available yet”	The project you are in has no models connected.	Switch project in the header, or ask an administrator to add models in AgentHub’s Model Manager.
”The assistant could not be reached (400).”	Most often the conversation has passed 20 exchanges, or the message is over 32,000 characters.	Start a New chat, or shorten the message.
”The assistant could not be reached (401).”	Your session expired.	Reload and sign in again.
”The assistant could not be reached (409).”	The project has no models.	As for “No models available yet”.
”This model’s API key is missing or revoked.”	No working provider key for that model.	Choose another model or Auto, and tell your administrator.
”The selected model is no longer configured.”	An administrator removed the model you picked.	Pick another model, or Auto.
”Model used its entire N-token budget before writing any answer…”	A reasoning model thought until it ran out of output tokens.	Pick a non-reasoning model for that question.
”Model returned no text…”	The provider returned an empty answer.	Send again, or try another model.
”A credential or government identifier was found in this message… send again to continue.”	The server found a critical item and held the message once.	Send the message again. It goes out masked.
Send stays grey	Empty message, an answer still streaming, or an attachment still Scanning….	Wait, or remove the stuck attachment.
The button says Send masked	A critical identifier (key, card, SSN and so on) was found.	Click it to confirm, or edit it out of your draft first.
The answer stopped partway	You clicked stop, or the connection dropped.	Ask again. A stopped answer may not be saved.
The view does not follow a streaming answer	You scrolled up.	Click New text below or Jump to latest.
An answer is marked Underpowered	The question scored harder than the model that ran.	Click Try model on the card to use a stronger model for your next message. If it says “No stronger model is set up here — ask your admin.”, none is connected.

Attachments

Symptom	Fix
”No selectable text in this PDF — it may be a scan.”	Run OCR first, or paste the text.
”This PDF is password-protected.”	Remove the password and attach again.
”Unsupported file type.”	Convert to `.txt`, `.md`, `.csv`, `.tsv`, `.json`, `.log`, `.pdf` or `.docx`.
”Files are limited to 10 MB.”	Split or trim the file.
”Too much text across attachments. Remove one first.”	The files together exceed 100,000 characters. Send them across several messages.
The chip says “first 50k of … chars”	Only the first 50,000 characters of that document are sent. Trim it to the part you need.

All attachment errors are listed in Attachments.

History and privacy

Symptom	Cause	Fix
A reopened chat shows `[PERSON_NAME_0042]` instead of a name	By design: personal details are removed before saving, and only the masked text is stored.	None needed. The banner “Reopened from history — personal details were removed before saving and show as placeholders.” explains it.
A key or card number shows as a placeholder even on my own message	Critical values are never shown again after sending, even to you.	Keep the original where you got it from.
A conversation is missing from the sidebar	History is per project and shows your 40 most recent conversations.	Switch to the project you used.
Chats show ”🔒 Encrypted chat”	End-to-end encryption is on and this browser is locked.	Click Locked — enter passphrase in the strip at the top of the chat.
I forgot my encryption passphrase	There is no reset.	Set a new passphrase for future chats. Past chats remain readable only by your organization’s super admins.
Something personal was not masked	Detection is not perfect, especially for names and addresses.	Edit it out before sending. Report the example to your administrator.
A product or tool name was masked as a person	Over-detection.	The model can usually still answer. Rephrase if not.

For administrators

Symptom	Cause	Fix
Employees see platform pages on ModelHub	The deployment was built with the wrong variant.	Rebuild ModelHub with `NEXT_PUBLIC_APP_VARIANT=user`.
A User sees “Your account has ModelHub access only”	They opened AgentHub.	Send them the ModelHub URL.
An employee sees the organization-wide usage dashboard	They are an Owner, Admin or super admin.	Change their role to User (Viewer) in User Management.
User Management says “Needs platform access first”	The email is not approved.	Approve it in Platform access, then save again.
An invited employee lands on “No models available yet”	Invite acceptance lands on the organization’s first project.	Connect models to the first project, or tell them to switch project.
Every message goes to the same model	Only one model is connected, or every connected model is the same size.	Connect a Small, a Mid and a Large model. See Rolling out ModelHub.
Conversations are titled with their first line instead of a short summary	The titling model on CorePillar’s platform key was unavailable.	Nothing to fix on your side; titles fall back automatically.

FAQ

Which model answered my question?

The model name is on the answer’s report card, and the routing line above the composer shows it while the answer streams. Details → Usage repeats it.

Can I always use the best model?

Yes. Pick it in the model picker and ModelHub will use it for every message until you change it. Routing never overrides your choice. Your report cards will tell you when a cheaper model would have done.

Does my company see my conversations?

There is no transcript viewer for administrators. Your organization’s Owners, Admins and super admins see your usage totals and, on the organization-wide usage dashboard, the types and values of personal data that were removed from your messages. Saved conversations are stored masked. If end-to-end encryption is on, your organization’s super admins can also decrypt your saved conversations.

Does the model provider see my personal data?

Not the data ModelHub detects. Detected items are replaced with placeholders before the message leaves CorePillar, including in earlier turns sent as context. Detection is not perfect, and confidential business text that is not a personal identifier is not masked.

Why is Optimise off?

It is off by default and resets when you reload or switch project. Turn it on in the composer if you want filler removed.

Can I rename, delete or search conversations?

Not yet. Start a new chat for a new topic; each conversation is titled automatically.

Can I upload images?

No. Attachments are text documents only: .txt, .md, .markdown, .csv, .tsv, .json, .log, .pdf and .docx.

Why does my usage include sources other than Chat?

AI Usage covers every CorePillar feature that called a model on your behalf, across all projects in the organization. If you also use AgentHub, your Cost Calculator runs and evaluation judge calls appear there too.

Is there a mobile app?

No. ModelHub is a web app used in the browser.

Get started

Using chat

Your account

How it works

For administrators

The chat

Attachments

History and privacy

For administrators

FAQ

​Sign-in

​The chat

​Attachments

​History and privacy

​For administrators

​FAQ

Sign-in

The chat

Attachments

History and privacy

For administrators

FAQ