Knowledge base
Let the model look things up in your own documents, through a LightRAG server you run.
The knowledge base is optional. It is backed by a LightRAG server that you run — on this computer or another — and Actovia only connects to it: it pushes your documents there and asks it for the passages relevant to a question. Your chat model always writes the answer.
Setting it up
- Run a LightRAG server (see its documentation).
- In Settings → Knowledge Base, enter its Server URL and, if you set
one, its API Key (your
LIGHTRAG_API_KEY). - Test Connection — it reports the server's embedding model and how many documents it holds.
Choose LightRAG's embedding model with care: once a document is in, changing it means wiping LightRAG's storage and adding every document again.
Query mode decides how LightRAG searches — Mix (the knowledge graph and vector search together, the best quality) is the default; Naive, Local, Global and Hybrid are also offered. Advanced holds Top K and Chunk Top K.
Documents
Documents on the same page lists what is in the knowledge base, each with its state: Pending, Indexing…, Indexed, Failed or Needs reindex. Add one with a title and its content; edit it and it is indexed again. Reindex all sends everything again.
In a chat
With a server set up, the Knowledge Base tool is on in Settings → Built-in Tools, and the model searches your documents when a question calls for it. There is one shared knowledge base for every chat and every Philharmonic group.