Atom
The smallest meaningful unit. One title, one idea, one body. The atom parser creates them automatically from every AI response.
Max 60 characters in the title. The body is the full original text.
Think in Atoms, Not Threads
You ask AI something. You get a wall of text.
Noema breaks every response into atomic units — one idea, one card, one tap.
No scroll. No lost context. Just the map of what you’re thinking.
The Problem
You ask AI something. You get 400 lines.
You scroll. You lose the thread. You ask again.
The answer was in there — buried under intros, follow-up questions nobody asked for, and three paragraphs of recap.
Noema ends the scroll.
How It Works
-
You type a prompt.
Choose your model — Claude, GPT, Gemini — and write. Noema sends the prompt and receives the raw response.
-
The atom parser runs.
A lightweight agent (Claude Haiku) decomposes the response into atoms: one concept per card, with hierarchy and parent relationships intact. No rephrasing. No interpretation. Structure only.
-
You navigate, not scroll.
Atoms appear on a spatial canvas. Tap to focus. The rest steps back — still there, just out of the way. One tap brings anything forward again.
The parser uses claude-haiku-4-5 at temperature 0 and returns strict JSON.
It never rewrites your content — it only imposes structure.
Core Concepts
Everything in Noema is built around a small set of ideas. Once you have them, the rest is obvious.
Canvas
A spatial workspace. Atoms are arranged by hierarchy, not by time. Root atoms at the top, children below. Drag anything anywhere.
No vertical scroll. Depth is in the layout, not in the thread.
Shadow
When you focus on an atom, everything else fades. Not deleted — suppressed visually. The data is always there.
Tap any shadowed area to bring it back immediately.
History
Every session is saved locally. Browse past conversations like a git log. Pin the ones you need, delete the ones you don’t.
SQLite on your machine. Nothing leaves your device.
Noise Filter
AI models add filler by default — intros, outros, disclaimers. Toggle noise off. Only signal atoms appear on the canvas.
Toggle back on anytime. It’s always your call, not the model’s.
Session
A session is a conversation with persistent context — like a chat, but the interface is a canvas. Start new, continue old, pin the important ones.
Context is capped per session to keep token usage predictable.
The Noise Filter
AI models are trained to be helpful in ways you didn’t ask for.
“Great question! Let me break this down for you.”
“I hope this helps. Let me know if you need anything else!”
“Here’s a summary of what we covered above…”
These are noise atoms. When the filter is on, the parser marks them and they’re hidden before they reach the canvas.
What’s left is only what the model actually said that was worth saying.
Works With Any AI
Noema doesn’t lock you to a provider.
- Claude (Anthropic) — Primary connector. Sonnet for conversation, Haiku for the atom parser.
- ChatGPT (OpenAI) — GPT-4o mini supported. Bring your own API key.
- Gemini (Google) — Gemini 2.0 Flash supported. Same interface, different model underneath.
The atom parser runs on Haiku regardless of which model you use for conversation. The parser is separate — cheap, fast, structured.
Read Anything. Atomically.
Open any article. Click the Noema icon. The page content is extracted, cleaned of navigation and ads, and sent to your canvas as atoms.
No copy-paste. No noise. Works on any website.
-
Open any article.
Navigate to whatever you want to read — news, docs, long reads, anything on the web.
-
Click the Noema icon.
The extension extracts the content — strips ads, nav, footers — and sends it straight to your canvas.
-
Navigate, not scroll.
Noema parses the content into atoms. You think through it, not past it.
- Readability.js — Mozilla’s open-source extractor strips chrome, keeps content.
-
Local endpoint —
The extension talks to Noema on
localhost:11420. Nothing goes through a server. - User-initiated always — Nothing runs in the background. You click, it runs, it stops.
Local-First
- SQLite on your machine — Sessions and atoms saved in a local database. No account, no sync, no cloud.
- API keys stay local — Keys configured in the app, stored in your environment. Not sent anywhere except the model endpoint.
- No telemetry — Noema doesn’t phone home. What you write is yours.
- Offline capable — The canvas, history, and past sessions are all available without a connection. Only model calls require network.
Built On
Rust (Tauri v2) backend · Vue 3 + Pinia frontend · SQLite (tauri-plugin-sql) · Atom parser: Claude Haiku, temperature 0, strict JSON.
Part of noema.tools
Noema is the knowledge and conversation layer of the noema.tools ecosystem. Pragma handles Claude Code sessions — structured, atomic, controlled. Same paradigm, different domain.
Get Noema
Windows and Linux. Local-first. No account required.