The organization portal

Your chat history, the shared knowledge base, what the assistant remembers about you and your own AI key — plus, for organization admins, channels, connections, the knowledge host and the licence.

Sign in at /org/login with the same email address and password you use in the CAD add-in. Your company name appears as a badge in the header. The sidebar has two sections:

Organization — Chat History, Knowledge, My Memory, My AI Key. Everyone sees these.
Admin — Messaging Channels, Connections, RAG Provider. These appear only if your membership role is org_admin; other members do not see them at all, and opening the URL directly shows a “this page requires an organization admin” message.

Sign-in needs a company membership, not just an account. If /org/login refuses you, ask your administrator to add your email to the organization — a registered account that belongs to no company cannot use this portal.

Chat History

Everyone Every completed CAD chat turn that was recorded, newest first. The scope the page is reading is shown next to the heading — your company, or your own account if you have no company scope.

What the controls do

User — organization admins pick All members or a single member. Ordinary members see their own name in a locked box: the filter is re-applied on the server for every request, so a member can only ever read their own turns.
Search in conversation text — a plain text match inside the stored turn. Type a word you remember using, such as a feature or part name.
Apply — runs the query. Results arrive 50 at a time; Load more appends the next 50.
Assistant reply — each card shows your message; click to expand the reply. The CAD system and the session identifier are shown as small labels.

Things that surprise people

A red banner saying history recording is currently disabled means an admin turned recording off on RAG Provider. Old turns still display; new ones are not saved.
If the knowledge service cannot be reached, the page shows the reason instead of an empty list — history lives in the same store as the knowledge base.
History is what makes the assistant able to recall facts from earlier sessions. Turning it off is a privacy decision with a capability cost.

Knowledge Base

Everyone uploads Admins delete Documents the assistant can search and cite in answers: design guidelines, standards, material data, part libraries.

Uploading

Choose a file. PDF, DOCX, XLSX, MD and TXT are accepted, up to 50 MB. A larger file is rejected before anything is sent.
Pick a Type: Document for reference material, Spec for requirements and standards. The type is also what the search box below searches, so keep it deliberate.
Give a Title if you want one; blank uses the file name.
Press Upload. The file is split into chunks and indexed; the confirmation tells you how many chunks were created.

Uploading the same content twice is detected — you get “already indexed (identical content)” and nothing changes. Rename the file all you like; it is the content that is compared.

Searching and managing

Search returns the five best-matching chunks with a relevance score and a snippet. It searches the Type currently selected in the upload row above — the most common reason a search finds nothing is that the document was indexed as a Spec while the search is running against Documents.
The Documents table lists everything indexed for the company, with the chunk count and who uploaded it.
Download returns the original file, and only appears when the original was stored. Delete removes the document from the index for everyone in the company. Both buttons are visible to organization admins only.

Users can also index a document straight from the CAD add-in — see Knowledge Base for that route and for how retrieval is used in an answer.

My Memory

Personal Short facts the assistant has remembered about you and uses as background in your chats — your preferred units, your usual material, a naming convention you follow.

To add one, simply tell the assistant in chat — for example “remember that I prefer dimensions in mm”. There is no add form on this page by design; memories come from conversation.
Forget deletes a fact immediately and permanently. Nobody else can see or delete your memories, and you cannot see anyone else's.
An empty list simply means nothing has been remembered yet.

My AI Key (bring your own key)

Personal Run your chats on your own AI provider account instead of the platform's. On a trial this is not optional — a trial user without a seat must supply a key before chat will run at all. A paid seat holder may add one to move their usage onto their own provider bill.

My AI Key page with provider, model, endpoint and API key fields — Org portal → My AI Key. The badge at the top shows whether a key is already on file and for which provider.

The fields

Provider (required) — the name that decides the wire protocol. Use anthropic for Claude, openai for OpenAI, groq for Groq, and openai-compatible for xAI/Grok, Gemini, DeepSeek, Mistral and most gateways.
Model (optional) — blank means “use whatever model the platform is configured for”. Fill it in to pin a specific model on your own account.
Endpoint (optional) — blank means the platform's endpoint. Required when you chose openai-compatible, because that is where the request goes.
API Key — pasted once. It is encrypted at rest, scoped to your user account, and never displayed again. When a key is already on file, leaving this box blank keeps the existing key — that is how you change only the model without re-pasting the secret.
Use my key for chats — the on/off switch. Unticking it keeps the record but stops it being used.

What happens when you save

The setting is live on your next chat turn; nothing needs restarting. From then on, the order the server uses to pick a key is your key → your organization's key → the platform key. Your key wins even if your company has one and even if you hold a paid seat. Tokens spent on your own key are billed by your provider and do not consume your plan's included allowance — the split is visible on the licence page.

Remove key deletes the record entirely and drops you back to the organization or platform key.

Creating a brand-new record with the key box empty is rejected with “An API key is required”. Blank only means “keep existing” once a key is actually on file.

Messaging Channels Org admin

Credentials that let your users chat with their CAD workstation from Telegram, Microsoft Teams or WhatsApp. All three channels are enabled by default for every company; a channel goes live as soon as its credentials are saved. Every secret is encrypted at rest, and a blank secret box on a channel that already has one means keep the stored value.

The last panel, User self-connect, decides whether people may link their own Telegram or WhatsApp account from the CAD add-in and the tray Agent. Turning it off disables those buttons and refuses link requests on the server; Teams still links automatically by work account.

After saving you are told that new Telegram bots are picked up when the messaging service next restarts — a brand-new bot token is the one setting that is not instant.

The full walkthrough for each channel, including the exact webhook endpoints and the BotFather steps, is on Channels & Messaging.

Connections Org admin

A connection is a named external system your users can drive from chat — “look up item 1000 in the ERP”. The credential is encrypted at rest and the call is made on the server, so the secret never reaches anyone's PC.

Connections page with the provider template picker, config JSON box and the list of saved connections — Org portal → Connections. Start from a template, replace the placeholders, store the secret once.

Start from an example fills in the type and a ready-made config for Dynamics 365 Business Central, SAP S/4HANA Cloud, ERPNext, Salesforce, HubSpot, Microsoft 365 mail, SendGrid, Postmark or a generic REST API, and suggests a name. A blue box tells you exactly which secret to paste and gives a first call to try.
Name is what people say in chat — keep it obvious: ERP, CRM, Email.
Config JSON holds the base URL and the authentication style. Saving is refused while a <placeholder> is still in the text, which catches the most common setup mistake before it becomes a runtime error.
Service-account secret is stored encrypted; on an existing connection a blank box keeps it.
Enabled takes a connection offline without losing the setup — prefer that to deleting.

Once saved, test it from chat straight away: ask CoPilot to list my connections, then to make one simple read. Provider-by-provider configuration with the exact JSON is on Connections (ERP & Email).

RAG Provider Org admin

Decides where your company's knowledge base and chat history are stored, and whether history is recorded at all.

RAG Provider page with the custom server checkbox, endpoint, API key and history toggle — Org portal → RAG Provider. Off means the shared platform service; on routes everything to your own server.

Use a custom (self-hosted) RAG server — when off, knowledge is stored on the platform's shared service. When on, documents and searches go only to the endpoint you give, so the content never leaves your infrastructure.
Endpoint and API key are only editable while the custom option is ticked. A green Set badge means a key is already stored; leaving the box blank keeps it.
Test connection calls the server and reports its embedding model, vector dimensions and version. Do this before saving — a wrong endpoint otherwise shows up later as an error on the Knowledge and History pages.
Store chat history in the knowledge base — when on, completed chat turns are indexed so the assistant can recall facts from earlier sessions. Turning it off stops new recording; anything already stored is removed from the Knowledge page.

Saving applies to the next request. Ticking the custom option without an endpoint is refused. Switching between the shared service and your own server does not migrate documents — whatever was indexed on the other side simply stops appearing until you switch back or re-upload.

Organization License

Dedicated deployments A read-only summary page for companies running their own dedicated CADCoPilot server, reached at /admin/org/license. It shows:

Licence status, plan and max users — an unlimited seat count is shown as an infinity symbol.
Contact — the name, email and phone recorded for the licence.
Dedicated server keys — each key your own server uses to authenticate, with its active/revoked state, how long ago it last sent a heartbeat and how many connections it reported. A heartbeat older than a couple of hours means the dedicated server is not talking to the licence server.
Usage (last 30 days) — total tokens and sessions, split into what ran on the plan's included allowance and what ran on your own AI key. The second figure is billed by your AI provider, not by us.

Nothing here is editable: licence changes and new server keys go through support.