Amalgama / Knowledge Readiness Check
Give your AI assistant better source material
Inspect duplicates, missing ownership and possible sources for employee questions. Turn a small document collection into a reviewed cleanup plan, entirely on your device.
Find the gaps before rollout
Start with the policies people actually use. Add their questions. Inspect the evidence together.
A fictional supplier policy with duplicate copies and a changed approval limit.
Source and ownership
Stored in this open deskDrop one TXT or Markdown file here
Remove this document
Removes this source and resets findings. Saved copies are unchanged.
Keep at least one document, or clear the collection below.
What will people ask?
One question per line. Optional for duplicate and ownership checks.
Add a name and text for every document to continue.
Find sources by meaning
Optional local search for English text. First use downloads about 23 MB of model weights plus runtime assets. No account, API key or document upload. Slower devices can keep using keyword checks.
Allow the asset download above to continue.
Build a reviewable cleanup plan
Run the collection check to see duplicate copies, missing ownership and source passages. Your original documents stay unchanged.
Clear this collection
Remove current text, questions and findings. Export anything you need first. Saved copies are unchanged.
Know what was checked
Upload TXT or Markdown, or paste text from a PDF. Up to 6 documents, 12,000 characters each and 24,000 in total. Up to 8 questions. No web crawling, permission audit or automatic factual verification.
Exact duplicates ignore case and whitespace. Version candidates share at least 72% of each document’s distinct words, with at least 8 words each. Keyword retrieval requires 40% of the question’s nontrivial words in a passage. English stop words are excluded.
Optional semantic search processes short passages with a local English-language model. It returns the nearest three passages even if none answers the question. This is retrieval, not a contradiction detector or an AI readiness certification. Review source context and record your conclusion.
Source text is not sent to a model server. Download hosts receive normal asset request metadata. On the website, export before closing. Workspace copies use your selected storage. Search model and component credits
Need this connected to your workflow? Discuss an integration