Explained
Chat with your own documents, privately
Connect your folders, PDFs, spreadsheets, and databases, and ask about them in plain words. Toskar finds the passages that answer each question and the answer cites them: local RAG, on your own computer, with nothing sent to an AI company.
What you can connect
Four kinds of source
Passwords and tokens for databases and web APIs are stored apart and never shown again.
A file or folder
Read in place on the computer Toskar runs on, and read again when the files change.
Upload or paste
Toskar keeps a copy of what you give it.
A database
One read-only SELECT query on a SQLite file, PostgreSQL, or MySQL. Each row becomes a passage, and Toskar never changes the database.
A web API
A URL that returns JSON, CSV, or text, fetched again when a question needs data older than the interval you choose.
How it works
Answers that show where they came from
When a chat's profile uses a source, the passages that match each question are added to the answer, and the answer cites them, down to the PDF page. Change the data and the next answer changes.
Step 1
Open Knowledge
Choose the kind of source under Connect knowledge, or start from an example.
Step 2
Choose what to connect
The file, folder, or database, pasted text, or a web address.
Step 3
Pick the profiles that use it
A chat looks in a source only when its profile uses it. The profile chats start with is checked already.
Step 4
Try a question
See which passages a question would bring into a chat, then ask it in Chat.
Connecting and changing sources takes an Admin account; everyone who signs in can ask questions that use them. See what your computer can run to pick a model.
What to get
The products that fit
Questions
- What is RAG?
- Retrieval-augmented generation. Before the model answers, the passages from your own files that match the question are found and added to what it reads, so the answer can use them and say where they came from. The model itself isn't changed.
- Which files work?
- PDF, Excel, CSV, TSV, JSON, JSONL, Markdown, text, and HTML. Each Excel sheet is a table, and each PDF page is cited by its number. Scanned PDFs are read with text recognition, installed on this computer the first time (about 110 MB).
- Do I need an embedding model?
- No, but it helps. Without one, search matches the words in a question. Install an embedding model, such as Nomic Embed Text v1.5 from Models, and it also matches meaning, so a question about a warranty finds a passage about a guarantee.
- Should I train a model on my documents instead?
- Usually not. Knowledge is for facts that change, such as prices, stock, and policies: change the data and the next answer changes, with nothing retrained. Training teaches a model a way of working, and when you build a specialized AI, Toskar recommends Training, Knowledge, or both for each file you add.
- Can my other apps search my documents?
- Yes. Apps that speak MCP, such as Claude Desktop, Claude Code, Cursor, and VS Code, can search your connected knowledge through Toskar, and the API uses it when a key allows.
- Do my files leave my computer?
- No. They're read and searched on the computer Toskar runs on, and models run on your own computers. Mark a source This computer only to keep every chat that uses it on this computer, even with other computers paired.