Private knowledge AI
Ask your documents anything. Share only what each answer needs.
NeuAira Lore reads your PDFs, Word files and scanned pages, then answers questions in plain language with a citation to the exact passage. Run it hosted by us, on your own servers, or fully offline on a network with no internet at all.
Early access Set up personally for each early customer.
Your team can't paste client files into ChatGPT. They still need answers.
Contracts, policies, case files and manuals hold the answers, but finding them means opening folder after folder. Public AI tools would help, if sending confidential documents to someone else's servers were an option. Lore gives you the same experience while you control what leaves: nothing at all on an offline appliance, or only masked, relevant passages in Lore Cloud.
What it does
Answers with sources
Answers stream in as they're written, each citing the document and passage it came from. Click a citation and the document opens at that page.
Understands tables, forms and charts
Reads PDF, Word, text and Markdown, including tables row by row, filled-in PDF form fields, and headers and footers. Charts and scanned pages are read page by page, so the figures in a chart can be found and quoted. Files up to 50 MB.
Finds the right passage
Combines keyword and meaning-based search, then re-ranks the results. Exact terms like a policy number and loosely worded questions both find the right passage.
Works with the numbers
Adds up, compares and calculates from the figures in your documents and cites where each one came from. It never brings in outside knowledge.
Ask about specific documents
Type @ to pick one or more documents and limit the question to them. Shorter documents are read in full.
Says when it doesn't know
Set a relevance threshold and Lore declines to answer when your documents don't support one, instead of guessing.
Privacy mode
Email addresses, phone numbers and other identifiers are swapped for placeholders before the language model sees them, then restored in the answer. Names can be masked too. Detection runs on the appliance, and identifiers in page images are blacked out.
Private by default
Documents are private to the person who uploaded them. An organisation can turn on sharing, but nothing ever crosses an organisation boundary.
Chat history that stays private
Each person's conversations are saved for them alone; admins can't read them. Each organisation chooses how long history is kept, or turns it off.
Connect your own storage
Index documents where they already live in Amazon S3 or S3-compatible storage. Access is read-only, nothing is copied in, and permission stays with you to revoke at any time.
Roles that make sense
Members ask and upload. Admins manage users, sources and API tokens. The operator who runs the server manages organisations and cannot see any of their content.
An API for your own tools
Upload, ask and retrieve passages over a documented API with service tokens, so Lore can power other apps and agents.
AI that runs on the appliance
In the Sovereign edition the language model runs locally on the appliance itself, so questions and answers never leave it. In Lore Cloud, a hosted AI model sees only the passages relevant to each question, with identifiers masked first.
Measured answer quality
A built-in evaluation harness scores retrieval and answers on a labelled set, or on your own documents and questions.
Simple to run
Installs as a self-contained appliance with everything it needs included. Updates apply in place, and backups are one command.
How it works
Everything in the diagram runs inside your boundary. In the offline setup there is no outbound connection at all, so every AI model ships inside the appliance rather than being downloaded.
One product, three ways to run it
The software is the same. Choose where it runs, and how much of it you want to operate yourself.
Lore Cloud
For firms that want private document AI without running a server.
- Hosted and run by us: nothing to install or maintain
- Documents stay in your own storage, such as an S3 bucket, and are read in place, read-only
- Only the passages relevant to each question are sent to a hosted AI model, never your document library; a short document may be sent in full
- Emails, phone numbers, account and ID numbers are masked by default before anything is sent, and restored in the answer
- Buy directly or through your MSP; priced per user, per month
Lore Sovereign
For a single firm that needs its documents fully in-house.
- Fully air-gapped or on-premises
- The language model runs locally on the appliance
- Offline licence key: nothing phones home
- Your own TLS certificate, or a private one for closed networks
- Licensed by number of users
Lore for MSPs
For managed service providers running AI for several clients.
- Many client organisations on one install, each kept separate
- Set each client's seats, document count and storage
- Monthly usage report per client, as a CSV for your billing
- Privacy mode and history retention set per client
- Each client can connect its own storage
- Host in your data centre or private cloud, or resell Lore Cloud and run nothing
- Licensed by organisations and users
Good fits
Law and accounting firms
Search contracts, engagement letters and precedents without sending client files to a third party.
Clinics and care providers
Find answers in policies, procedures and protocols, with staff only seeing what they're allowed to.
Manufacturers and engineering firms
Ask questions of manuals, specifications and quality records that are too sensitive to leave the site.
Defence and government suppliers
Run on a network with no internet access at all.
Questions about Lore
What hardware do we need?
A single server you provide, or one we specify with you. Everything, including the language model, runs on that hardware, and a graphics card makes answers much faster. We size it with you during setup.
How is it licensed?
By a signed licence key that's checked offline, so it works on air-gapped networks. It sets how many organisations and users you can create. If a subscription lapses, Lore keeps answering questions; only adding new organisations pauses until renewal.
Can we evaluate it first?
Yes. Without a licence, Lore runs as a single organisation for evaluation. Early-access customers get a guided trial on their own documents.
Can it read documents where they already are?
Yes, for Amazon S3 and S3-compatible storage: Lore indexes the files in place with read-only access, and an admin must accept a plain statement of what that allows before connecting. You can also upload files or send them through the API. SharePoint and Google Drive are on the roadmap; tell us which ones matter to you.
Can it answer questions about tables and charts?
Yes. Tables are read row by row with their headers, and pages with charts are described when they're uploaded, so a question like "what was revenue in Q1?" can be answered from a bar chart and cited to its page.
Try Lore on your own documents.
Early-access customers get a guided setup, founding pricing and direct access to the engineers who build Lore.