Newshot Explore any topic with AI-powered mind maps
Generate a structured overview, follow marked terms into connected explanations and keep every branch on an infinite canvas.Built to keep model usage costs low.
Free and source-available. Bring your own API keys for cloud or local models. Projects stay in your browser.
Interactive demo
The demo is simplified on touch screens. The app is designed for desktop browsers with a mouse or trackpad. Here the demo opens full screen with fewer controls.
The demo replays recorded answers for this project. Click a marked term to open it, or type in a follow-up field or the chat to ask a prepared question. Nothing is sent to a model provider; reload the page to reset the demo.
The demo replays recorded answers for this project. Tap a marked term to open it, hold a term for its preview, or type in a follow-up field or the chat to ask a prepared question. Nothing is sent to a model provider; reload the page to reset the demo.
Features
Start with a topic, a chat or an image
Generate a structured overview of a topic, start a conversation, or upload or link an image to ask about it. Add your own sources to the project at any point.
Hover for an explanation, click to explore
The app marks terms worth exploring in generated answers. In encyclopedic nodes, expandable bullets suggest directions for further exploration. Hovering a term shows a short explanation. Clicking it creates a connected node, or takes you to the node that already covers it.
Add images, GIFs and videos
Upload, paste, drop or link images and GIFs. Add video files or YouTube and Vimeo embeds. Media nodes sit in the map alongside text and interactive nodes.
Ask about a node, a passage or part of an image
Follow-up questions can refer to a whole node, selected text or selected image regions. Questions about image regions build an atlas-like map with the image node at the center.
Run interactive nodes on the canvas
Generate an interactive node from a prompt or from an existing explanation. Run it on the canvas or in an enlarged window. Request a change to the program, or ask a follow-up question without modifying it.
Arrange branches and groups
A physics engine spaces connected nodes apart, and pinned nodes stay where you put them. Select several nodes to move, pin, group or delete them together. Groups hold single nodes, selections or whole branches, with manual or automatic hierarchical colors.
Answer from your own sources
Add PDFs, documents, webpages, images or pasted text. Restrict answers to your selected sources only, or combine them with the model’s knowledge and, when enabled, web search. Citations let you inspect the referenced material.
Easy to read and understand
Information chunking gives each sentence one core idea on its own line. Key parts are bolded or marked as terms. A reading ruler follows the text line by line or word by word. The default font stays legible at small sizes; OpenDyslexic and other fonts are available in settings. If the ASCII background is distracting, lower its opacity in Appearance settings.
Also in the app
- Node typesExpandable encyclopedic nodes, focused explanations, chats, editable Markdown text nodes, media nodes and interactive nodes.
- ChatsBranching, file attachments, conversation history, fact-checking individual answers, copying an answer to a separate text node, and more.
- Web searchThrough supported providers or a configured search service, with citation previews.
- Fact-checkingCheck text sentence by sentence. Review cited evidence where available, see which claims lack support, and rewrite the text using the evidence.
- Markdown editorVisual and source editing with tables, tasks, code, mathematics and images. Export as Markdown or Word.
- Image generation and editingGenerate images or use AI to edit the whole image or a selected region. Crop, adjust, draw or annotate with text manually.
- Free imagesAutomatic or manual search through Wikimedia Commons and Openverse, with attribution and license information.
- ProjectsProjects save automatically. Search for nodes or switch to an earlier version of a node or chat answer. Import and export entire projects. Back up and restore your app settings separately.
- Obsidian CanvasExport the project as a canvas with editable Markdown notes, media and connections. Keep the layout when you open it in your vault.
- SynchronizationProjects and settings through Google Drive or WebDAV, including self-hosted Nextcloud, with snapshot history.
- Multiple AI modelsSwitch between several main models on the canvas. Assign other models by task and control reasoning, token limits and background activity.
- Deep customizationAdjust appearance with colors, fonts and canvas effects. Change behavior with keyboard shortcuts, reading controls and layout settings.
AI costs
While your model choices and provider rates determine the bill, the app is designed to keep costs low. You choose which models handle which tasks, set output token limits, and decide whether to generate term previews automatically. The recommended models balance answer quality, speed and price.
- Use powerful models where accuracy mattersYour main model writes node explanations and chat replies, where factual knowledge and a low hallucination rate matter most. Generating a text answer typically takes fewer tokens than writing and testing code. For a new text node, the app sends at most the parent and grandparent nodes as canvas context—not the whole project.
- Keep routine work inexpensiveA separate, cheaper secondary model marks terms, writes previews and generates titles. Repeated background tasks need not run at the main model’s rate.
- Limit reasoning by defaultMain and secondary tasks default to low reasoning where the model supports it. This limits the tokens and time spent thinking; you can raise the main model’s effort when a question warrants it.
Necessary exceptions
- Interactive nodesThe coding model writes, tests and revises a working program. Passing code and test results between steps uses more tokens than a text answer, but gives the model a chance to catch and fix mistakes before you use the node.
- Image generationFor educational images, we recommend a higher-quality model, even if it costs more. Look for one that draws labels, proportions and relationships accurately instead of inventing details. Even a convincing image can be wrong; check important details against a source.
Your projects stay off our servers
Projects, sources, saved media and settings are stored in your browser. No separate app account is required. Your browser sends requests directly to your model providers; no app server or API relay sits in between. API keys stay on your device by default. Syncing them is opt-in. We do not receive or store your projects, prompts, answers or API keys on an app server.
Read the privacy notice
- Projects and sources
- Settings and API keys
No app server between your browser and these services
Get started
Connect a model provider or a model server running on your computer. With OpenRouter quick setup, you don’t need an API key in advance.
- Open the app in a desktop browser.
- Use quick setup for OpenRouter with recommended models, or custom setup to add providers and choose models yourself.
- Set a spending limit with your provider where supported, and use a separate key for this app if possible.
- Start with a topic, a chat or an image.
Generation, term previews and other background tasks can incur charges with your provider. AI answers, citations, generated images and fact-check results can be wrong.
Providers and models
Connect OpenRouter, OpenAI or Anthropic, or use Ollama, LM Studio or an OpenAI-compatible endpoint. Gemini models are available through OpenRouter. Local servers must allow browser connections from the app’s address.
| Main models | Explanations, conversations, image questions and fact-checking |
|---|---|
| Secondary model | Term marking, previews, titles and other background tasks |
| Interactive model | Creating and modifying interactive nodes |
| Image model, optional | Generating and editing images |
One model can serve several roles. Unassigned secondary and interactive roles fall back to the main model.
Run it locally
A model server such as Ollama or LM Studio runs AI models on your computer. The app sends requests to it directly from your browser, so they never leave your computer. Local models need no provider account and have no usage fees. Speed and answer quality depend on your hardware and the model.
Each role can use a model from a different provider, so local and cloud models can work together. For example, a local secondary model can mark terms and write previews in the background while a cloud main model writes the explanations. Image generation needs a cloud image model.
The app itself can also run on your computer from its source code, independent of this website. With the app and the models both local, explanations and chats work without an internet connection; web search, free images, embedded videos and sync still need one.
For local servers, the app’s provider settings include a connection guide. Step-by-step instructions for running the app are on GitHub.
Source-available
The app’s original code is published under the PolyForm Noncommercial License 1.0.0. It permits use, modification and redistribution for the purposes the license defines; commercial uses outside those permissions need separate permission.
This makes the app source-available, not open source under the Open Source Definition. Third-party components and assets keep their own licenses.
Support me
I’m Jakub Fleites-Jonczyk. I’m originally from Poland and now live in Germany, where I’m in my fifth year of psychiatry residency. In my spare time, I build web apps and other things. I have ADHD and built this app so I could study in a way that works for me.
My residency takes up most of my week. If you find the app useful, a tip on Ko-fi could help me spend more time working on it—or at least cover the hosting bill.
Opens Ko-fi in a new tab for one-time or monthly tips. Privacy details.
Contact me
For questions or bug reports, email [email protected] or open a GitHub issue.