madebyzan/knovum

Bring your knowledge together. Make it useful.
A local-first AI workspace for research, learning, and creation.

Download Knovum · What you can do · Privacy · FAQ · Build from source · 简体中文

What is Knovum?

Knovum is a local-first desktop workspace for gathering, exploring, and creating with your own knowledge.

Bring your notes, saved webpages and videos, meeting recordings, PDFs, Office documents, images, and code into one visual space. Organize them on an infinite canvas, discover connections, search across your library, and ask questions with answers that link back to the source passage, page, or timestamp.

Turn the material you have collected into documents and media without losing the trail back to the sources behind them.

Your library lives on your machine. There is no Knovum account or Knovum cloud. Search, transcription, OCR, and other supported AI capabilities can run locally; when you choose a cloud model for an answer or generation task, Knovum sends only the context needed for that task to the provider you selected.

What you can do

• Bring everything into one place. Collect notes, webpages, videos, recordings, PDFs, Office files, images, and code in a Space.
• See how ideas connect. Arrange material on an infinite canvas, connect claims with evidence, and keep related questions and decisions close to their sources.
• Ask questions and check the answer. Search by keyword or meaning, ask a connected AI model about your material, and follow citations back to the exact source location.
• Create from what you know. Turn selected material into documents, images, videos, subtitles, or voice output, with a record of the inputs and model used.
• Keep control of your data and AI. Store your library locally, use supported on-device capabilities, and choose the model provider or compatible desktop Agent for each task.

How it works

_{From scattered sources to connected knowledge and work you can trace.}

1. Collect. Paste a link to save a webpage or a YouTube, Bilibili, TikTok, or Vimeo video. Add recordings, PDFs, Office files, images, code, and notes. Recordings can be transcribed locally with timestamps and speaker labels.
2. Organize. Arrange material in sections and groups. Add claims, evidence, questions, and decisions as knowledge objects, then connect them with meaningful relationships.
3. Explore. Search one Space or your whole library by keyword or meaning. Ask questions about your sources and follow each citation to its original location.
4. Create. Select the material you need and create a document, image, video, or voice track. Export your work while keeping a record of the sources and revisions behind it.

Explore the features

Spaces: one canvas for all kinds of material

Rich notes, Markdown, webpages, PDFs, Office documents, images, audio, video, code, and ordinary files share one infinite canvas. Organize content with sections, groups, tags, and Saved Views; navigate with an outline; and undo the last 100 edits. A system Inbox gives new material a place to land before you organize it.

Knovum supports the open JSON Canvas format (.canvas) for exchanging canvases with compatible tools. You can also move a complete Space — including its content, assets, relations, derivations, and citations — as a portable .knovumspace package.

Ask questions and follow the sources

Ask questions about your own material with a connected model. Answers cite the passages they rely on, and each citation keeps its location in the source: a page in a PDF, a time range in a recording, or a heading in a note.

Citations are tied to the source revision they quote. When a source changes, a citation can become detached instead of silently pointing to content that has moved. A PDF citation opens the original in the built-in reader at the relevant page, with the passage highlighted.

Find information across your library

Hybrid keyword and vector search runs on your machine using Qwen3 embeddings and LanceDB. Search within one Space or across all Spaces, and find related material by meaning as well as exact terms.

Connect ideas into structured knowledge

Use six built-in knowledge object types — Concept, Question, Claim, Evidence, Experiment, and Decision — or create your own. Connect them with relationships such as supports, contradicts, answers, depends on, supersedes, and derives from.

Knowledge checks can surface unsupported or contested claims, evidence without grounding, answered questions, and notes whose sources have changed. You can use as much or as little structure as your work needs.

Create from selection, with a record of where it came from

Turn a selection into a new resource without losing its context. Create documents, generate images or videos, remove backgrounds, split image layers, or add subtitles. Each run records the engine, model, inputs, and input revisions, so you can trace a result back to the material used to create it.

Work with documents, media, and voice

• Editors: Rich notes and Markdown with tables, images, find and replace, and inline citations.
• Export: Notes and captures to DOCX, PDF, PPTX, HTML, Markdown, and TXT; image and audio conversion.
• Video workbench: An absolute-time timeline, proxy playback, and burned-in subtitles.
• Speech studio: Audiobooks with roles and a pronunciation lexicon, narration, dubbing, vocal/accompaniment separation, voice cloning, and voice design.
• System-wide dictation: Hold a global shortcut, speak, and release. Whisper transcribes locally and inserts the text into the app currently in front.

Connect compatible AI agents

Knovum includes a Builtin Agent that can plan with todos, research your Space read-only, and make changes through validated, atomic change sets. You can also use desktop Agents already installed on your machine through the Agent Client Protocol (ACP), with their own sign-in and tools.

Knovum exposes your Space to compatible desktop Agents through a loopback-only MCP server, and can connect outbound MCP servers as Agent extensions. Skills, plugins, connectors, and apps can be managed in one place.

Local-first by design

Your knowledge library is stored on your device: Spaces, files, extracted text, search indexes, and backups stay in local application data. Knovum does not require an account and does not run a Knovum cloud service.

The distinction between local and cloud AI is explicit:

• Local by default for supported capabilities. Editing, keyword and semantic search, transcription, speaker recognition, OCR, speech synthesis, and the listed image tools run on your machine after their models are downloaded.
• Your choice for answers and generation. Connect a provider you already use, run a compatible local endpoint, or hand a task to a desktop Agent.
• Only the task context is sent to a cloud provider. When you choose a cloud model, the action sends the selection and context it needs to the provider you chose. Provider-side handling is governed by that provider's terms and settings.

Model files are downloaded only when a feature needs them, verified against pinned SHA-256 digests, and can be removed from Settings. Download progress and cancellation are available in the Task Center.

Use your own AI provider

Canvas editing, keyword and semantic search, transcription, and the on-device capabilities below work without an AI account.

For AI answers and generation, connect a provider you already use:

OpenAI · Anthropic · Google Gemini · DeepSeek · Qwen · Zhipu · Moonshot · Doubao · MiniMax · StepFun · SiliconFlow · Mistral · Groq · Grok · Together · OpenRouter · Kling · Ollama · any OpenAI-compatible endpoint

You can also hand a task to a desktop Agent that brings its own credentials. Keys are stored in a dedicated local secrets database and are never written to logs.

Download and get started

Download the latest release

Builds are not yet notarized by Apple or Authenticode-signed, so macOS and Windows ask you to confirm the first launch. Updates after that are delivered in the app and verified against the release signature. Linux and Windows on ARM are not supported yet.

A simple way to start: create a Space, add two or three sources you are already working with, and write down one question you want to answer. Add an AI connection in Settings when you want AI-generated answers or content.

FAQ

echo 'export HF_ENDPOINT="https://hf-mirror.com"' >> ~/.zshrc
source ~/.zshrc

On-device AI capabilities

The following capabilities run locally. Models are downloaded on first use, verified against pinned SHA-256 digests, and can be removed from Settings.

Embeddings and transcription use Metal on Apple Silicon. None of the capabilities above requires Ollama or another model server.

Build from source

You need Node.js 22.13+ with Corepack, the Rust toolchain pinned in rust-toolchain.toml, and the native tooling used by the desktop build: uv, CMake, Protobuf, and LLVM/libclang. See docs/installation.md for exact versions.

git clone https://github.com/madebyzan/knovum.git
cd knovum
corepack enable
pnpm install --frozen-lockfile
pnpm desktop:dev

The first build compiles the native inference worker and takes longer; later starts are faster.

CI runs these checks on macOS (Apple Silicon and Intel) and Windows. Pushing a v* tag builds the macOS and Windows packages into a draft GitHub Release.

Architecture

A Rust core owns data, inference, and OS integration; a Vue 3 renderer owns interaction; generated contracts connect the two. Local models run in a separate inference worker, and the Builtin Agent runs in a bundled Python runtime.

Tauri 2 · Rust · Vue 3 · TypeScript · SQLite (sqlx) · LanceDB · Candle · ONNX Runtime · whisper.cpp · Konva · Tailwind CSS v4

Read more in docs/system-architecture.md, the documentation map, and the product design contract in DESIGN.md.

Contributing

Bug reports, focused feature proposals, translations, and documentation fixes are welcome. See CONTRIBUTING.md and the Code of Conduct.

When opening an issue, include your platform, Knovum version, and steps to reproduce. Remove credentials from logs. Report security issues privately as described in SECURITY.md.

If Knovum is useful to you, a ⭐ helps other people find it.

License

Knovum is released under the MIT License. Third-party components retain their own licenses.

Star history