Private AI for business: an on-premise LLM that answers from your documents
Corpus is private AI for your documents: an open-weight language model with search over your files, and nothing goes to a public AI API. Set up for you: it runs on your own servers or on a cloud you choose. You ask in plain words; it answers from your files with a citation to the document and page. If nothing on file supports an answer, it is designed to say “Nothing on file.”
Private files should not go to a public AI service.
Public AI chat tools send your question, and any file you paste, to someone else’s model. For contracts, client files and project records, many organizations will not accept that.
What teams often say:
- “We can’t paste client files into a public chatbot.”
- “IT wants to know where the model runs and who can read the prompts.”
- “We tried a chatbot. It sounded sure of itself, and nobody could check where the answer came from.”
- The files sit on network drives, SharePoint and scanned PDFs, and the people who own them want them to stay where they are.
How a private deployment is set up
- Set up for you Where it runs: a local NVIDIA GPU on your network, our GPU server, or a cloud you choose (AWS, Azure or Google Cloud). We set it up with you.
- The model: an open-weight model, run on that server. Corpus uses no public AI API.
- Set up for you What runs where: we work from a copy of your files on that server, and search and the model run there too. Your originals are not touched.
- What your IT reviews: where the server sits, who can reach it, which role sees which files, and how each copy is kept and deleted (see data handling).
- How it answers on that server: search over your files, with each answer cited to the document and page. Enterprise RAG walks through it.
Example questions, not results
Written for this page; the demo’s own files are engineering documents.
- Which of our past contracts tie the limitation of liability to the fees paid?
- What does our policy manual say about approving spending over the limit?
- Which project reports mention groundwater at the site, and on which page?
- What changed between the first and the final version of this report?
Corpus finds and cites. People decide.
- A person reads the cited page before relying on an answer.
- Whether the server, who can reach it and how it is backed up meet your own rules.
- Which people get which role. Your people decide; Corpus applies the roles they set.
- Your own security review. We don’t have a SOC 2 report, which is one reason you can choose your own servers.
Corpus is not a general chatbot. It answers from the files you load, and it is designed to say “Nothing on file” instead of guessing.
A free call, then a pilot on your own files.
Start with a free 30-minute call. Then a $5,000 fixed-price pilot on a copy of one real project’s files. How the pilot works.
See how citations and refusals work.
Try page citations, “Nothing on file” and roles on a sample engineering firm’s files.
→ Corpus for engineering firms → 7 RAG refusals that were pipeline bugs → Enterprise RAG → Corpus by industryFrequently asked questions
What is a private LLM?
A large language model that runs on infrastructure you control instead of a public AI service. Set up for you Corpus uses an open-weight model run that way and adds search over your files, so each answer cites the document and page.
Can Corpus run fully on-premise?
Yes. Set up for you Corpus is cloud-agnostic. It runs on AWS, Azure, Google Cloud, or your own servers, for example a local NVIDIA GPU on your network.
How is this different from a public AI chatbot?
With a public chatbot, what you send goes to the vendor’s servers. Corpus answers only from the files you load, cites the document and page for each answer, says “Nothing on file” when nothing supports one, and uses no public AI API.
Is there a demo?
Yes. The live demo is built for engineering firms: a fictional sample firm with real public records from US public agencies. Its search also reaches a public-domain collection of U.S. court opinions. There is no demo built on another industry’s files. A pilot runs Corpus on a copy of your own files.
Where do our files go?
Set up for you Corpus is cloud-agnostic. It runs on AWS, Azure, Google Cloud, or your own servers. It uses an open-weight model, and nothing goes to a public AI API. In the pilot, your files are never used to train or fine-tune a model. In full deployment, fine-tuning on your files happens only if you agree in writing. We don't have a SOC 2 report, which is one reason you can choose your own servers.
Have a folder full of questions?
Book a free 30-minute call. We talk about your files, the questions you need answered, and whether a pilot makes sense.