Two ways to answer
Classic mode matches a question to the best prepared answer, fast and predictable. Agent mode reasons across turns with tools: search answers, search services, ask for clarification, remember details and log gaps.
Knowledge that improves every week
Your documents become audited answers the persona can stand behind. Whatever it can't answer becomes a gap your team closes once, for every surface at the same time.
Animation: documents break into question-and-answer cards that feed a search index. Two questions arrive and match their answers. A third finds nothing, becomes a knowledge gap, is answered once and joins the index.
From documents to answers
Four steps, run in the background as soon as you upload.
- 1
Upload
PDF, Word, Excel or plain text. Scanned PDFs go through OCR, and an optional vision pass reads tables and layouts page by page.
- 2
Generate
The engine detects the language, splits the content, and writes question-and-answer pairs, in Arabic, English or both. Near-duplicates are removed.
- 3
Audit
Each pair is rated for clarity. Weak ones are rewritten automatically, and your team can filter, edit or remove any pair.
- 4
Index
Answers are embedded into a search index of their own for each knowledge base, so one persona never sees another's content.
The knowledge-gap loop
Every assistant meets questions its content doesn't cover. What matters is what happens next.
- 1
Detected, not guessed
When the closest answer isn't close enough, the persona doesn't improvise. It logs the question, the language, why nothing matched and what topic it seems to belong to, with personal details removed first.
- 2
Grouped and ranked
Repeats of the same question are merged and counted, so the gaps your customers hit most rise to the top.
- 3
Drafted for you
One click drafts an answer in Arabic and English, with alternative phrasings, into a suggestions queue. Or write the answer yourself.
- 4
Approved and live
Approving creates the answer, indexes it immediately and, if the persona has a face, renders the answer clip. The gap is marked closed.
How the persona uses what it knows
Asks before it assumes
When it isn't sure what someone means, the persona asks a clarifying question. If uncertainty continues, it moves the conversation to a stronger model.
Fast on repeat questions
Embeddings and intent results are cached at the edge, so common questions are answered without repeating work.
Built-in manners
Greetings, thanks, goodbyes, out-of-scope questions and “let me check” moments are handled by intent responses written to match the persona's gender and your organization's name.
Send us three documents
We'll turn them into a working persona and show you the first gaps it finds.