What is ModelHub

ModelHub is the AI assistant your company gives to every employee. You open modelhub.corepillar.ai, type a question, and get an answer from one of the models your administrators have connected. In the app it introduces itself as CorePillar Assistant. It looks like an ordinary chat app. The difference is what happens between pressing Send and the model reading your message:

​

Who it is for

• Employees who need a capable AI assistant without choosing between a dozen models or worrying about pasting the wrong thing.
• Administrators who want every AI conversation in the company to run on approved models, on the company’s own provider keys, with spend and privacy events visible per person.

ModelHub is a separate deployment of the same platform that powers AgentHub. Employees get the chat; administrators configure models and review usage in AgentHub.

​

What you can do in ModelHub

• Chat with streaming answers rendered as Markdown, with syntax-highlighted code blocks and copy buttons.
• Let Auto pick a model, or choose a specific connected model from the picker.
• Attach up to five text documents (.txt, .md, .csv, .tsv, .json, .log, .pdf, .docx) per message. They are scanned and masked like your typed text.
• Reopen past conversations from the sidebar.
• See your own spend, token use, model mix and privacy events on the AI Usage page.
• Change your profile photo, switch between light and dark mode, and switch organization or project.

​

What ModelHub does not do

Being clear about limits saves you time:

• No model setup. You cannot add models or API keys in ModelHub. If you see “No models available yet”, an administrator has to connect a model in AgentHub’s Model Manager.
• No other platform pages. Traces, evaluations, the cost calculator and the rest of AgentHub are not reachable from ModelHub. Typing their URLs sends you back to the chat.
• No conversation rename, delete or search. The sidebar lists your recent conversations; there are no controls to manage them yet.
• No regenerate or edit. To change a question, send a new message.
• No images or audio. Attachments are text documents only. Scanned PDFs with no selectable text are rejected.
• Reopened chats show placeholders. Personal details are removed before a conversation is saved, so when you reopen it later you see [PERSON_NAME_0042] rather than the name. This is by design.
• Not a general confidentiality filter. Screening finds identifiers and credentials. It does not recognise confidential business content such as unreleased figures or strategy.

​

Next steps