Start with the manual you already have. Connect data later — or never.
Every business runs on documents nobody reads twice: standard operating procedures, onboarding manuals, the how-we-do-things pages someone wrote carefully and everyone asks a colleague about instead. Give those documents to Sonela and your staff can ask them questions in plain language, inside the application they already work in — and get answers the way your business does things, not the way software generally works.
Paste a chapter in, or pick a .md or .txt file and your browser reads the text out of it. Each document holds up to 100,000 characters — a real manual chapter. Anything longer is refused with the exact count, never silently cut, because a silent cut would leave you believing the assistant was told something it will never read.
The assistant reads your documents in the order you set, first to last, so you decide what matters most. A workspace holds a bounded set of chapters rather than a bottomless archive — and when a single answer cannot carry everything, the assistant is told exactly what it was not shown, so it can say the manual covers this instead of inventing the missing chapter.
The same one line that installs every Sonela assistant. Drop it into your ERP, CRM or internal tool and the assistant appears where your staff already are — nothing else about your application changes, and if you later connect live data it is the same widget, the same workspace, the same key.
A Sonela workspace can run on your written material and nothing else — no database connected, nothing of your infrastructure touched. That is not a degraded mode; it is the first rung of the same ladder every workspace stands on.
On guidance alone, the assistant knows what it cannot see. Ask it for a count, a total or a live status and it says, in one sentence, that no database is connected here and that connecting one would unlock answers from the live data — then helps with whatever part your manual does cover. It will not invent, estimate or recall a value as if it had read one.
When you want live answers, the same workspace climbs: connect your data through the Sonela Gateway — one small service on one server inside your network, dialling out over port 443 — and everything you set up stays set up: the widget key, the branding, and every chapter you uploaded.
The workspace takes two kinds of written material, and they do different work. The grounding context is vocabulary: one note, up to 4,000 characters, read before every question — what your status codes mean, which numbers people actually ask about, what a word means in your business. SOP documents are procedure: the chapters the assistant answers from, up to 100,000 characters each, in your order. Put what words mean here in the context and how we do things here in the documents, and each question gets both.
The same candour as everywhere else on this site: here is where the edges are. The assistant also stays in its lane — questions that are not about your application and your material are declined briefly, with a pointer to something it can help with instead.
On guidance alone there are no counts, totals, statuses or documents to read, and no tool to query for them. The assistant says so — one sentence, then back to what your manual covers. Live answers arrive when you connect data, not before.
Paste, or a .md or .txt file, is the whole of v1. Reliable PDF extraction needs layout and table handling we have not built, and a bad extractor would silently feed the assistant garbage you believe it read — so PDF support ships when it works, not sooner.
Uploaded procedures are stored in your workspace — that is what makes the answers possible — and go to the AI provider with the question they help answer. Only your workspace can read them; deleting a document, or the workspace, removes it. Our usage records carry sizes and counts only, never the text.
Chapters reach the model after the rules that govern it, framed explicitly as background data with none of the rules' authority. A sentence in a document that tries to change them is ignored, and the rest of the chapter still answers questions.
The SOP library is not a tier or an add-on: every Sonela workspace carries one, whatever the plan. The free trial — 14 days or 200 answered questions, whichever comes first, no card — is enough to load your manual and let your team ask it real questions, and if the trial ends before you decide, the workspace pauses with every chapter preserved. Paid plans differ in connected data sources and included answered questions, not in whether the assistant reads your manual: the current prices are here.
Yes — that is what the SOP library is for. Upload your written procedures and the assistant answers "how do we do this here" questions from them, inside the application your staff already work in, the way your business does things rather than the way software generally works. When your manual does not cover a question, it says so instead of guessing.
No. A workspace can run on your written guidance alone, with no database connected at all. On that footing the assistant knows it cannot see live records: ask it for a count or a status and it says, in one sentence, that no database is connected here and that connecting one would unlock answers from the live data — then helps with whatever part your manual does cover. Connect data later through the Sonela Gateway, or never; it is the same workspace either way.
It is an assistant inside your own business application — your ERP, CRM or internal tool — added with one script tag, for the staff who work there. It is not a public marketing-site chatbot: each widget key answers only to the sites you list for it, and the assistant declines questions that are not about your application and your material.
Text and markdown today: paste a chapter in, or pick a .md or .txt file and your browser reads the text out of it. PDF is deliberately not supported yet — extracting text from real PDFs reliably needs layout and table handling we have not built, and a bad extractor would silently feed the assistant garbage you believe it read. When PDF support ships, it will ship working.
They are kept in your workspace — that is what makes the answers possible — and they are sent to the AI provider with the question they help answer: the provider you chose if you brought your own key, the one Sonela runs if you did not. Only your workspace can read them, and deleting a document, or the workspace, removes it. Our usage records carry sizes and counts only, never the text.
Open a workspace, paste in a chapter, and ask it a question — before you have connected anything at all.
14 days, 200 answered questions, no card. No AI key to supply — trials run on Sonela's key.
