How Hebno works

Most search tools work on keywords and folders. If you don’t know what a file is called or where someone saved it, you don’t find it. Hebno works from meaning and from the client the file belongs to, so you can ask “what did we agree with Contoso on late delivery?” and get the clause, the deal it belongs to and the invoices that followed. Hebno does this in four steps that run on their own every time a file is added or changed. A fifth rule applies to numbers, which Hebno never copies.

​

1. Hebno reads everything you point it at

Connectors read from the places where your firm stores its knowledge: SharePoint and OneDrive, Google Drive, Dropbox, Box, Confluence, Notion, your project files and the workspace library. When a file is added or changed, it is read within minutes. Nobody uploads, moves or renames anything. Hebno reads a document the way a person does, not as a stream of text. Headings remain headings, a table remains a table with its rows and columns intact, a scanned contract is read from the page image, and a slide is read with its title, its chart and its speaker notes. A pricing table in a proposal is therefore searchable as a pricing table, not as a jumble of numbers. Admins choose which sites, drives, libraries and spaces Hebno reads, and can exclude folders such as HR or partner-only areas. See Organisation brain.

​

2. Hebno labels each document

A model reads each document the way a new joiner would and records what it is: This replaces the tagging and filing nobody does anyway. It also marks which version counts: when a folder contains SOW_v3.docx, SOW_v3_final.docx and SOW_v3_final_signed.pdf, the current one is marked and answers use it, while the older drafts remain findable when you ask for them.

​

3. Hebno indexes by meaning, not only by words

Each document is split into short passages, and every passage is stored with the context it came from: the document, the section heading and the labels above. Each passage is stored as a mathematical fingerprint of what it means, which is the “vector” part. A search for “penalties for late delivery” finds a clause titled “liquidated damages”, even though no word matches. A normal keyword index runs alongside it, so exact terms still hit: a contract number, a product code, a person’s name. Hebno searches both and ranks the results together, then reads the best passages in full before it answers.

​

4. Hebno links documents to your business records

Because each contract is labelled with its client, Hebno connects a Contoso contract to Contoso’s deals in your CRM, its jobs in your ERP and its invoices in your finance system. Search stops being one big pile of files and becomes a map of each client: the documents, the people, the open work and the money, side by side. To build this map, only names and identifiers from your business systems are stored, such as a client name, a deal number or a job code. Hebno does not copy the records themselves. Linking a project to its records on the Brain tab makes the connection explicit. See Linked records.

​

Numbers remain in their source systems

Hebno does not copy financial or delivery figures into its index. When someone asks about margin, budget burn, hours logged or what a client owes, Hebno looks it up live in the system that owns the figure, at the moment of the question. You always get today’s figure, every number is cited to the record it came from, and any calculation shows its inputs. The model never produces a figure on its own. The same applies to anything personal or fast moving. Email, calendars, meeting transcripts and chat messages are searched live in each member’s own account when a request needs them, and are never added to the shared index.

​

Access rules match the source

If someone cannot open a file in SharePoint today, they will not see it in Hebno either. Hebno records the permissions of every document it indexes and checks them twice: once when it searches, and again against the member’s own access when it reads the document to answer. When access is removed in the source system, or a file is deleted, Hebno updates within minutes. The same applies to business systems. Live lookups run with the member’s own account or with the team and data-area rules your admins set, so financial figures are visible only to the people allowed to see them. See Permissions and access.

​

Security

• Your region: your workspace and its index are stored in the EU or the US, as you choose. Admins can turn on only EU-hosted models when work must remain in the EU.
• Names are masked: names of clients, people, projects and amounts are swapped for placeholders before text reaches a model, and restored inside your workspace when the answer comes back.
• No training: your inputs and outputs are never used to train a model.
• Encryption: AES-256 at rest and TLS 1.2 or higher in transit.

Read more on the Hebno security page.

​

From a question to client work

The sources searched depend on your question: the index, a live system, the web, or several at once. A quick question gets a direct answer. Work that needs evidence from many places becomes a research run, with each step and source shown as it goes. Either way, every reply lists its Sources, and deliverables mark each finding as Supported, Inferred or Uncertain, so a person reviews the work before it reaches a client. Hebno writes back to another system only after you confirm a card.

​

Next steps