OleeASK answers questions from your own documents - grounded and cited - with department-level access control. Then it checks your people actually know them.
Not just boxes and text. Here is each part of OleeASK, shown the way you will use it.
Ask in plain English and get an answer drawn only from the documents you are allowed to see - with the exact source quoted. If it is not in your documents, OleeASK tells you instead of inventing one.
Full-time staff receive 21 days of paid annual leave per year, accrued monthly.
HR Handbook §4.2We couldn't find this in the stores you can access - so we say so instead of guessing.
Stores for HR, IT, Finance and beyond, with nested folders and a clean document browser. Uploads are OCR’d, chunked and embedded automatically until each file is ready to answer.
Pick source files, describe what you need, and get a formatted document back - headings, lists and tables intact. Export to Word, PDF or Markdown in a click.
Drop in two versions and get an inline redline down to the phrase, with an AI summary of what moved. Keep A, keep B, take both or remove - then merge and download the final, in an editor that matches the original.
Refunds are issued within 30 days of purchase.
Refunds are issued within 14 days of purchase, minus a processing fee.
Build exams straight from your material. Multiple-choice grades instantly; written answers are scored by AI against your own documents, with feedback you can see.
Refund window correct and the faulty-goods exception is cited. Missed the change-of-mind processing fee.
Grant "can ask" and "can upload" per store, to a person or a whole team at once. Departments stay isolated, C-level can see everything, and every change is staged until you save - and written to the audit log.
Admins see questions over time, the activity mix, most-active members and score distributions. Members get their own trend and marks. The platform owner sees usage only - never your content.
Most tools summarize from a few retrieved snippets and quietly miss the rest. OleeASK reads the entire document once at upload, distills it into a grounded coverage summary, and reuses that everywhere - so executive summaries are instant and assessments are drawn from all of the material, not just page one.
At upload we read the entire file end to end - not a handful of retrieved snippets. Nothing important is skipped because it sat past the first few pages.
A grounded model distills the full text into a faithful coverage summary the moment indexing finishes. It is saved, so the executive summary is instant and included - never recomputed.
Every claim must trace to the document. The model is instructed to use only what is on the page - no outside knowledge, no invented names, numbers or dates.
Questions are generated from that whole-document coverage, so an exam spans all the material - beginning to end - instead of testing only the opening section.
We stuff the full text into a long-context model (with map-reduce for very large files), so coverage is complete - the known failure mode of retrieval-only summaries is avoided.
Analyses cite the source and refuse to guess. If the material does not support a point, it says so rather than inventing one.
The executive summary and assessment source come from the stored summary, so they are consistent, instant and included - you only pay per page for extra on-demand analyses.
Bulk uploads run through a durable queue, not a best-effort background task. A big folder indexes steadily and safely in the background - capped so it never overwhelms the AI, retried when something transient fails, and watched so nothing gets stuck. It is built in and adds no token cost of its own.
Drop a folder of hundreds or thousands of documents and each one is recorded as a durable job - not a fire-and-forget task that vanishes if a tab closes or a server restarts.
A global concurrency cap means only a handful index at once, so a huge batch never overloads the AI provider or trips its rate limits - it drains steadily in the background.
A transient hiccup (a rate limit, a network blip) is retried a few times with increasing delays. Only genuinely unprocessable files - like a scan needing OCR - are marked failed.
A scheduled check re-runs any job interrupted mid-flight, so nothing is silently left half-done. You always see live per-document status, and can retry a failure in one click.
Drop documents into stores. OCR, chunking and embedding run automatically until each store is ready to answer.
Add your people, group them into teams, and choose exactly who can ask which stores - one person or a whole team at once.
Staff ask in plain English and get cited answers. You run exams to confirm they actually know the material.
Your documents - answerable, access-controlled and auditable. OleeASK is invite-only: enter your code and your workspace opens straight away. No code yet? Talk to us.