Browse help topics
- Help centre
- AI Chat Assistant
AI Chat Assistant
Use the built-in AI assistant to analyse documents, answer questions, and streamline your workflow.
The AI Chat Assistant gives you two separate ways to work with Accupe’s built-in AI, which is powered by Mistral AI. The AI Assistant side panel is always available and always answers the same way. The full AI Chat page adds the features that need more control.
On the AI Chat page you can upload documents so answers come from their actual content rather than from general knowledge, and you can pick a response mode, Fast, Planning or Ultra, depending on how much thinking a question deserves. Docs Only mode goes further and forces every claim back to a source in the documents you uploaded, with citations attached. Rate limits apply across both surfaces to keep responses fast for everyone, and the product tells you when you are close to one.
Two surfaces, and knowing which one you are in
The AI Assistant panel opens from the sparkles icon in the top bar on any page, and always answers the same way: one model, no mode selector, no document upload. The AI Chat page, reached from AI Chat in the sidebar, is the fuller surface, adding document upload, the response mode selector, Docs Only mode and a scope selector. One quirk worth knowing: choosing “Ask AI assistant” in the command palette takes you to the AI Chat page rather than opening the panel. Only the sparkles icon opens the panel.
Availability differs too. The panel is on every plan; the AI Chat page is Growth and Elite only, with Solo shown an upgrade prompt in place of the chat interface. Inside the panel a badge tells you which model is answering: Basic on Solo, Advanced (built on Mistral Large) on Growth and Elite. You do not choose that, your plan does. Both surfaces are for staff accounts only, and neither is available to a client signed in to the portal.
Neither is a general chatbot. Both are connected to tools that read your firm’s live data, so the AI can check upcoming deadlines, summarise your workload, look up a client, search jobs and tasks, look up a UK company on Companies House or verify a UAE TRN against the FTA register, calculate tax figures, and draft a client message. It can also list the historical AML screenings still awaiting a disposition, though it cannot run a new screening, because that capability was withdrawn from the product. Anything with a real-world effect, such as sending a message or applying a workflow template, asks you to confirm before it acts.
Documents, response modes and Docs Only
Upload happens only on the AI Chat page. The upload filter accepts a wide list of types, but accepting a file is not the same as reading it. Modern .docx and .xlsx files have their text extracted; PDFs are read directly when they carry a text layer and put through OCR on the first three pages when they do not; images always go through OCR. Older binary .doc and .xls files, and PowerPoint in either format, are accepted on upload but no text is pulled from them at all. They land in the document tray marked “No text” and the AI cannot read them, so convert to PDF, .docx or .xlsx first. OCR currently recognises English only.
The mode selector sits above the chat box. Fast gives a short direct answer. Planning lays out an Analysis Plan before the answer, which is useful when you will have to explain the reasoning to somebody else. Ultra runs a deeper multi-pass process structured as Analysis Plan, Deep Research, Verification and Conclusion, and takes noticeably longer. Fast and Planning are usable by anyone who can open the page; Ultra is Elite only, and clicking a locked mode sends nothing, it opens an upgrade card instead. Alongside it, a scope selector covers Company, All Clients and Specific Client, and the two settings combine.
Docs Only mode forces every answer back to your uploaded documents. Accupe then runs the reply through two automatic checks before it reaches you: that citation markers are present and cover most of the answer (more than 40 per cent of factual sentences arriving uncited fails), and that the key facts, names and figures genuinely appear in your uploaded content. A failure withholds the answer rather than risk an ungrounded claim. Citations render as small superscript numbers with a Sources line naming each document, and that numbering follows retrieval order rather than upload order once a session holds more than one document, which is why the Sources line is the definitive mapping.
Three separate limits, and how each is shown
The daily prompt quota is 10 on Solo, 50 on Growth and unlimited on Elite. It is tracked per team member rather than pooled across the firm, shared across both surfaces, and reset at UTC midnight. A prompt counts only once the AI finishes a full turn and gives you a final answer, so a question that makes the AI check several things behind the scenes still costs one. On the AI Chat page a usage meter shows your standing, for example “7 / 50 prompts today”, with a resets-in countdown, refreshing roughly every 30 seconds.
Separately, chat requests are throttled at 10 within any 60 second window, and that one is applied per network address rather than per login, so colleagues on the same office network or VPN exit share it. Document uploads have their own throttle of 10 per 60 seconds, tracked per user account instead, because processing a file is heavier than sending a message. And one ceiling sits underneath all of them: any single chat message can draw on at most 10 documents, whatever your plan.
What people come here to do
- Pick the right AI surface. Panel against page, Basic against Advanced, and what the assistant can actually read from your firm.
- Ask questions about a document. Supported types, which of them Accupe genuinely extracts text from, and how the storage counting works.
- Choose Fast, Planning or Ultra. What changes in the answer structure, which plan unlocks each, and how scope combines with mode.
- Force answers to cite your documents. The two verification checks that run before you see a reply, and what to do when they keep failing.
- Understand the limits you are hitting. Daily quota, burst throttle, upload throttle, and what a lapsed subscription does to AI access.
5 articles in this category
Things that catch people out
A file that uploads is not always a file the AI can read
PowerPoint files and older binary .doc and .xls files are accepted and then marked “No text” in the document tray. If Docs Only keeps failing, check that badge before you rephrase anything: no amount of rewording will fix a document that produced no text.
The usage meter and the throttle measure different things
You can be well inside your daily prompt quota and still see “Too many requests” because you, or somebody else on your network, sent several questions within the same few seconds. Wait a moment rather than assuming your allowance has gone.
Elite is unlimited on documents but not per message
The 10-documents-per-message ceiling applies on every plan. On Growth it rarely bites, but on Elite a session that grows past 10 uploads will fail every further message until you delete some back down or start a new chat. Split a large batch across several chats.
A very large document is rejected after processing, not before
If the text extracted from a single file exceeds 500,000 characters the upload is refused at the end of processing. A long wait followed by a rejection is expected behaviour rather than a fault.
A lapsed subscription stops AI before it stops much else
A cancelled or unpaid subscription blocks AI Chat with a re-subscribe message, and a payment failing for more than 14 days asks you to update your card. Inside that 14 day window you are still let through, so a temporarily declined card does not cut you off immediately.
Where to go next
- Documents & Vault. The separate AI Data Extraction tool in Company Documents, and why it behaves differently on PDFs than on images.
- Billing & Subscriptions. Daily prompt allowance, document limits, OCR level and access to Ultra mode are all set by your plan.
- Compliance & KYC. The assistant can list historical AML screenings awaiting a disposition, but running a screening is not something Accupe does.