Self-hosted · OpenAI-compatible
A research notebook that thinks with your sources.
Drop in documents, links, or pasted text. Ask questions. Get answers grounded in your sources, with citations. Bring your own model — any OpenAI-compatible embedding or chat endpoint.
—Notebooks
—Sources
—Indexed chunks
—Active chat model
Notebooks
Notebooks
List, create, rename, and delete notebooks. Search across titles and previews.
Notebook
Sources panel with drag-and-drop, file upload, URL fetch, and pasted text. Chat window with citation-aware answers.
Settings
Pick an embedding model and a chat model. Set per-model base URL and API key for any OpenAI-compatible provider.
How a question gets answered
Grounded retrieval, end to end.
Each message is embedded, used to find the six most relevant source chunks in this notebook, and sent to the chat model alongside your sources. The answer references them inline.
- You add sources: drag a PDF, upload a file, paste a URL, or type a note.
- Each source is split into chunks and embedded with the configured embedding model.
- You ask a question. It is embedded the same way.
- The top k = 6 chunks are pulled by cosine similarity and added to the chat prompt as
<source>blocks. - The chat model answers, citing sources as
[1],[2], … which link back to the right source.