Currently AI models only have a very limited context size and also they forget after the session ended.
But with specially crafted models adding the ability to understand and execute function calls its possible to add memory to the model and also to retrieve knowledge from large sets of data. In this article we outlined some approaches and core differences.
Using a set of parsers, and some loader functions (see list at the end) it is possible to utilize the CMS front-end as user interface for your knowledge-management AI.
InitBox offers an integrated solution based on MemGPT or Cheshire Cat via an open source CMS to „chat“ with your attached documents and wiki content.
Admin features via integrated settings interface:
- MemGPT /Cheshire Cat
- Select AI backend, eg. Oobabooga/llama.cpp or ollama
- Select LLM
- Switch to limit results to local knowledge by group, category and permission
- Select category to add
- Select max hardware usage / % of GPU/CPU resources
- Select vector db (defaults to qdrant) for embedding
User features:
- Upload via cms to file gallery or wiki
- convert to wiki
- Add to group
- Edit permission
- Approve correct conversion eg of graphs & tables in pdf or that content conversion is accurate
- Define revision depths for results
- Show/link original wiki/pdf page
- Reinforcement cycle to weight results
- Switch to limit results to local knowledge
- Export to XML
- Parse XML to AI
- Chat via Plugin Chat from any wiki page
- Chat with topic specific AI agent
- Use user language as default and add auto translate option
Enhance your knowledge-management with state of the art technologies.
Offering a well established user experience, fine grained permission control and full fledged work-group functionality.
Get in contact if you want to know how. We are looking for early adapters.
There is a lot to learn.
Some input formats available right now:
- „.csv“: (CSVLoader, {}),
- # „.docx“: (Docx2txtLoader, {}),
- „.doc“: (UnstructuredWordDocumentLoader, {}),
- „.docx“: (UnstructuredWordDocumentLoader, {}),
- „.enex“: (EverNoteLoader, {}),
- „.eml“: (MyElmLoader, {}),
- „.epub“: (UnstructuredEPubLoader, {}),
- „.html“: (UnstructuredHTMLLoader, {}),
- „.md“: (UnstructuredMarkdownLoader, {}),
- „.odt“: (UnstructuredODTLoader, {}),
- „.pdf“: (PyMuPDFLoader, {}),“.ppt“: (UnstructuredPowerPointLoader, {}),
- „.pptx“: (UnstructuredPowerPointLoader, {}),
- „.txt“: (TextLoader, {„encoding“: „utf8“}),