Answers with a source, or none at all
Every answer cites the documents it comes from. If it finds nothing relevant, it doesn't call the model: it tells you it doesn't know.
On top: artificial intelligence
Ask your company's documents and get the answer with its source.
Arkvia's chatbot answers questions about your company's documents. It always cites the document and page, respects what each person is allowed to see and, if it can't find the answer, it says so.
How it's used
Seven phases in the real app: you decide what it can read, ask questions and check the source, share the conversation, work in a space with your team, create assistants for repetitive work, and manage it all.
Animated demo of Arkvia's chatbot, in seven phases. Phase 1: in the knowledge base, the documents from document management that the chatbot may read are selected in one go; the confidential budget never goes in; then the specifications and the basic design for a tender are uploaded to the chatbot and processed in a few seconds. Phase 2: the question is narrowed down to one client and asks about the completion period and the penalty for delay; the answer cites the contract and the licence, with their pages; to a question that isn't covered by the documents, it replies that it can't answer it, and the conversation is starred. Phase 3: the citation opens the contract in the viewer, on page 7, where the clause is; from the document, the chatbot is opened alongside it and asked about the guarantee. Phase 4: the conversation is shared with the foreman, who can take part, and with the site manager, who can only read it; the foreman asks a question and gets an answer based on what he can see, and the site manager duplicates the conversation to carry on by himself. Phase 5: a workspace is created for the tender, with two people and the two documents, and a team conversation answers only from what's in the workspace. Phase 6: an assistant that summarises contracts is created from an example, with the contracts filter; its answer is saved as a PDF in the project and the assistant is passed on to another person, who receives a copy. Phase 7: in administration, usage and cost are reviewed, a person is given access and a rates document is blocked for everyone who uses the chatbot without being an administrator.
Features
Every answer cites the documents it comes from. If it finds nothing relevant, it doesn't call the model: it tells you it doesn't know.
PDF, Word, PowerPoint, Excel, CSV, text, images and BC3 (Presto, Arquímedes). Images and diagrams in PDFs are described using computer vision.
Quantity and budget figures are read from the file, never calculated by the model. If something can't be read reliably, it says so.
It only uses what that person can already open. Workspaces group projects and documents without ever granting extra access.
Saved instructions for repeated tasks, personal or for the whole team. The result is saved as a file (Markdown, Excel or PDF) in the project, with its sources.
Private, favourite and shared conversations. It works on mobile and, if the connection drops, it keeps the question and picks up the answer where it left off.
FAQ
No. It only answers with passages from your documents and cites the source. If it can't find relevant information, it doesn't call the model and tells you it doesn't know.
Yes. It reads BC3 (Presto, Arquímedes), Excel and CSV. Quantities are taken from the processed data, never calculated by the model, and if a file can't be read reliably, it says so.
Exactly what they can open in document management. The chatbot uses the permissions by client, project and document, and confidential documents are never indexed.
Fixed instructions that each person (or the administrator, for the whole team) saves for repeated tasks, such as reviewing a health and safety plan or extracting line items. Their result can be saved as a file in the project.
One infrastructure
They share documents and permissions: more productivity today and the foundations for artificial intelligence.