Skip to main content

What Tusab is

Tusab is a personal knowledge management (PKM) system with local AI. Point it at what you want to learn — a YouTube channel, PDFs, documents, notes — and Tusab absorbs it all, indexes it, and answers your questions in natural language, always citing the exact source the answer came from.

Runs entirely on your machine. Works offline with Ollama. Zero cost in the default setup.

The name

Tusab (Ancient Egyptian: s-bȝ-yt) is the name of a literary genre from Ancient Egypt — instructional texts, where someone of great experience recorded what they learned for those who came after. The most famous example is the Maxims of Ptahhotep (c. 2400 BCE).

The product does the same thing, in practice: you accumulate knowledge, Tusab organizes and indexes it, and when you have a question, it instructs you back from what you chose to teach it — never inventing, always citing the source.

The core pipeline — IAC

StageWhat it does
I — IndexExtracts and indexes any source: YouTube, PDFs, DOCX, Markdown, pasted text, audio, images, WhatsApp, meeting transcripts
A — AugmentRAG (Retrieval-Augmented Generation): retrieves the most relevant excerpts from your base and delivers them to the model as context
C — ChatNatural-language conversation, streaming responses, always citing title, date, and link of the original source

Why it exists

The central problem of personal knowledge management: you save a lot, but can't find it when you need it. Keyword search is poor, you don't remember where you stored something, and end up re-reading everything from scratch.

Tusab solves this with a mentor that only knows what you chose to teach it — which isn't a limitation, it's the guarantee that it won't hallucinate beyond what you indexed.

The four profiles

Tusab adapts to whoever is using it. The profile is chosen during onboarding and can be switched at any time.

ProfileMain use
🎓 StudentImports ready-made bases (.tusab) shared by teachers. Query only — zero setup.
📚 TeacherExtracts YouTube channels, indexes teaching material, exports bases to share with the class.
🔬 ResearcherBuilds corpora from multiple sources, uses Broad Search with semantic reranking, accesses project analytics.
🧑‍💻 SpecialistFull access: system monitoring, administration, full reset, all advanced tools.

What changes between profiles: visible tabs, the assistant's default persona, access to AI provider configuration, the extraction queue, Drive integration, and administrative tools.

What does not change: RAG quality, local-first privacy, mandatory source citation, and zero cost with Ollama — those are product invariants, regardless of profile.

What sets it apart

  • YouTube channel extraction at scale — entire channels, hundreds of videos, via local yt-dlp
  • Multi-source — YouTube, PDF, DOCX, Markdown, free text, images (OCR), audio (local transcription), WhatsApp, meeting transcripts (Zoom/Teams/Otter)
  • 100% local — data never leaves the machine; works offline with Ollama
  • Mandatory source citation — every answer cites title, date, and origin link
  • Teacher → student flow — export a ready .tusab base, the student imports it and starts chatting right away, no extraction needed
  • BYOK — Groq (free), OpenAI, Anthropic, Google Gemini as optional providers, alongside local Ollama
  • MCP Server — expose your knowledge base to Claude Code, Cursor, or any MCP client
  • Windows and macOS (Apple Silicon) — native installers for both systems

Next steps