[1]Connect your data, ask questions, and get answers that link back to the source.
Sources that connect to Critique Chat
Every quoted passage is matched back against your file before you see it. Documents are processed securely and never used to train shared models. PDF, DOCX, TXT, MD, CSV — up to 25MB. For persistent libraries and public/private isolation, open the full demo.
Try it yourself, see the citations, then hand it over. We connect the docs with you and run everything from there.
Upload a PDF, Word file, or plain text and ask questions about it right away — no signup, no setup. Text is extracted server-side the moment you drop a file in.
Every reply quotes the passage it used and points to a page or section in your file, so you can open the source and read the wording yourself.
One working session to connect your documents and systems. From there we build, host, tune, and maintain the assistant — you get a dashboard with usage analytics and nothing to operate.
Helpfulness isn't a bigger prompt. It's knowing how your sources connect, and spending compute where it counts.
They stuff chunks into a prompt, re-read the same context, and still miss how your sources relate. Fluent, but not useful.
Critique agents build a structured mapping between your documents and the answers they produce, so replies stay grounded, cited, and specific.
We don't burn tokens re-reading context, so more of the budget goes into reasoning over your knowledge instead.
Nothing to run or maintain. Your dashboard shows what was asked and cited, while we handle indexing, tuning, and upkeep.
Often you should. They are excellent general assistants and we use them every day. The difference shows up the moment someone has to defend the answer — to a client, a regulator, a reviewer, or a board.
A general assistant: Whichever passage scored closest to your wording. If retrieval doesn't fire — and it often doesn't — the model answers from general knowledge and reads exactly the same.
Critique Chat: Your documents, with the passage it used quoted directly in the reply. If the answer isn't in your files, it says so instead of filling the gap.
A general assistant: A citation names a fragment retrieval handed over — often without a page, section, or enough quote to locate it in the original file.
Critique Chat: Each reply quotes the passage and cites a page or section number, so you can flip straight to the source instead of hunting through the document.
A general assistant: People with a seat, each working from their own uploads. Shared access and role boundaries are on you to invent.
Critique Chat: One connected library, with assistants scoped to what each audience is allowed to see — your team keeps the fuller answers.
A general assistant: Someone re-uploads the file. The superseded copy usually stays where it was, and keeps getting cited.
Critique Chat: Connected at the source, so the answer moves when the document does. The old version stops being an answer.
A general assistant: You do — the uploads, the prompt discipline, the permissions, the re-indexing, forever.
Critique Chat: We do. One working session to connect your systems, then a dashboard showing what was asked and what it cited.
Where they win: drafting, brainstorming, code, and anything that doesn't have to be traced to a document. We are not trying to replace that. Put a document into the demo and watch it check its own quotes — that is the part they don't do.
We run the system, you keep the ownership. Retention, access, and keys are set once with you — then it just works.
Your credentials never leave your side of the gate. Nothing sensitive reaches the browser, so your team uses Critique without handing over the kingdom.
Demo files live only for your session. In production you set retention and storage once, and we enforce it. Nothing you share trains a shared model.
Your team can share one library without exposing everything to everyone. We'll match the solution to the way your organization already works.
Each reply quotes the passage it used and points to a page or section in your document. Trust comes from being able to open the source, not from the assistant's tone.
You are billed on tokens, not seats — so the whole organization can ask questions without you buying a license for each person. We connect and index your sources, run the deployment, and meter one all-in rate.
From $3
per million tokens — billed monthly
One metered rate on your own connected sources. No seat licenses, no minimum, no charge for the work we do to get you live.
Custom terms
Volume rates and annual agreements
Fully managed and procurement-ready. We integrate, host, and maintain Critique where your staff — or the public — already work.
A million tokens is roughly 20 to 25 fully cited answers, depending on how deep the assistant has to read. Tokens cover the entire answer — reading your documents, reasoning, and returning sources — so there is nothing metered on top.
