Knowledge Base
Findings You Can Search by Meaning
Papers, protocols, Study Reports and results in one Knowledge Base, with hybrid search that runs on your own server.
Hybrid Search
Keyword matching, including Chinese characters, fused with semantic vectors by reciprocal rank fusion, with an optional rerank. Live
Vectors on Your Server
Qdrant runs on your own AI server and is the single search store. Install needs Internet access once. Live
Public and Private
Organisation documents and personal ones, with Projects as tags. Names are renamable per organisation.
AI Tags
A summary, type and tags for every document, so folders stay searchable.
Folder Upload
A desktop app uploads whole folders with subfolders, and a vision model reads scanned drawings.
Chemistry Lab Feeds It
- Study Reports are saved to Public, Reports, Lab in the Knowledge Base
- Verified sources keep their DOI, tier and retraction status
- File storage can be a server, an S3-compatible store or the web server
- Computation files from a Study are kept in a per-organisation store with a typed-confirmation purge
Knowledge Base Questions
How does search work?
Documents are split into 1200-character chunks and embedded. Qdrant, running on your own AI server, stores the vectors. Search combines keyword matching with semantic vectors using reciprocal rank fusion, with an optional rerank, and rights are checked twice.
Who can see a document?
Public documents belong to the organisation and Private documents belong to one person. Projects tag folders and documents, and access rights are enforced on every search.
Can it read drawings and scanned PDFs?
Yes. A vision model can read PDFs that have no text layer. A desktop Knowledge Base app uploads whole folders with their subfolders.
Where are my files stored?
On a paired server, an S3-compatible store (AWS S3, Cloudflare R2, Backblaze B2, Wasabi, MinIO) or the web server, as you choose. Cloud vector databases have been retired; Qdrant on your own server is the single store for search.
Run Your Next Molecule Through AiMolSim
Sign in, describe a molecule and the question you have, and approve the plan the assistant proposes. Or request a demo first.