Nodepad: AI canvas that thinks beside you, not for you

You know that feeling when you're researching a topic and your ChatGPT conversation spirals into a wall of text you can't visually navigate? Every AI tool gives you a sequential chat — ask, answer, ask again — but your actual thinking is spatial and associative, not linear. You end up with 40 tabs, a messy notes app, and no clear picture of how your ideas connect. Nodepad addresses this by putting your notes on a canvas where the AI works in the background to classify, connect, and synthesize them without dominating the interaction.

You type a note into an input bar, and it lands on a spatial canvas. The AI immediately classifies it into one of 14 types — claim, question, idea, task, entity, quote, reference, definition, opinion, reflection, narrative, comparison, thesis, or general — and adds a short annotation that contributes something the note doesn't already say. As you add more notes, the AI infers connections between them based on content. When enough notes accumulate, a synthesis sentence emerges that bridges the tensions across your canvas. You can view everything in three layouts: a tiling grid, a kanban board grouped by type, or a force-directed graph showing centrality. Your data stays in localStorage and your API key goes straight to the provider — nothing passes through Nodepad's servers.

• 14-type auto-classification — your notes get labeled as claim, question, idea, task, entity, quote, reference, definition, opinion, reflection, narrative, comparison, thesis, or general without you doing anything, so yo...
• Emergent synthesis — when enough notes accumulate, the AI generates a single sentence that bridges tensions across your entire canvas, giving you an insight you might have missed
• Three spatial views — tiling grid, kanban board by type, or force-directed graph — so you can switch between structured and exploratory thinking without losing context
• Zero-server architecture — your API key and data never touch Nodepad infrastructure; everything runs in your browser via localStorage and direct provider API calls
• Multi-provider support — works with OpenRouter (including free-tier models at ~200 req/day), OpenAI direct, or Z.ai, so you pick your cost and quality tradeoff
• Export and import — save your canvas as Markdown or a versioned JSON (.nodepad) file, and import .nodepad files to resume or share projects

Who It Is For

If you're a researcher, writer, or product manager who thinks visually and finds chat-based AI tools confining for brainstorming, this is built for you. If you need local model support, collaboration, or mobile access, wait — all 9 open GitHub issues are requests for exactly those features, and none are implemented yet.

Worth Exploring?

Worth a look if you want a privacy-first AI research canvas that runs locally in 60 seconds. The maturity is experimental — no releases, no CI, 2 contributors, 76 commits — so treat it as a promising design experiment, not production infrastructure. The architecture is clean and the concept is genuinely different from chat-based AI tools, but localStorage-only persistence and the ~5MB browser cap mean it's best for individual research sessions, not long-term knowledge bases.