Drop files here

or click to browse — PDF, DOCX, PPTX, CSV, images, audio, video, ZIP

Pull from live server

Not connected

Authenticated with the token configured on the server (API_TOKEN).

Auth format Max records

Different server? Override the URLs

List endpoint Detail endpoint

Transcribe audio & video Downloads each course's media and runs speech-to-text so the spoken content becomes searchable. Without this only the media URL is indexed, not what is said. Much slower — transcripts are cached, so a re-pull skips finished files.

Indexed documents

Nothing indexed yet.

Paste credentials once. They live in the server's memory until it restarts — nothing is written to .env or disk. Leave a field blank to keep whatever .env already provides.

Language model

Provider LLM_PROVIDER Ollama model LLM

Ollama URL LLM_URL

Groq API key GROQ_API_KEY Groq model GROQ_MODEL

Context sizing — auto-detected, override only if needed

Left blank, the app reads the model's real context window and sizes the prompt to fit a full top_k=20 result. Fill these in only to force a value — e.g. lower LLM_NUM_CTX if a big local model is exhausting RAM.

Ollama window LLM_NUM_CTX Context chars PROMPT_MAX_CONTEXT_CHARS

Answer length LLM_MAX_TOKENS Groq tokens/request LLM_MAX_REQUEST_TOKENS

Groq's free tier meters tokens per minute (12,000 for llama-3.3-70b), which is far below the model's 128k context window — sending a full window returns HTTP 413. Raise this on a paid tier.

Database

User DB_USER Password DB_PASSWORD

Host DB_HOST Database DB_NAME

Named DB_HOST (not HOST) on purpose — this app already uses HOST for its own bind address.

Ask a question

Answers are generated only from the documents you have uploaded. Press Enter to send, Shift+Enter for a new line.

Chunks to retrieve Smart mode