Neuronotes 2.0

Local-first desktop MVP for notes, projects, and shared memory between humans and AI agents.

The app is intentionally Notion-like: quiet project sidebar, focused Markdown notes, project tasks, visual note map, compact context export, and visible agent provenance. It is not a chatbot-first product.

Documentation

• Deep project documentation: docs/PROJECT_DOCUMENTATION.md
• Routes and filesystem paths: docs/ROUTES_AND_PATHS.md
• Agent operating rules: AGENTS.md

Stack

• Desktop: Electron
• Frontend: React, Vite, TypeScript
• Backend: FastAPI
• Storage: SQLite
• Search: basic SQLite LIKE search across notes, tasks, and decisions
• Local model: Qwen 4B through configurable Ollama or LM Studio endpoint
• MCP scaffold: backend/app/mcp_server.py

MVP Scope

Included:

• Three main modes: Notes, Map, LLM
• Notes-first interface with a cleaner Notion-like writing surface
• Light/dark mode toggle persisted locally
• Create/edit projects
• Create/edit Markdown notes
• Improve note action powered by local Qwen with deterministic local fallback
• Project tasks that can be created manually or extracted from improved notes
• Visual memory map for notes, tasks, and relations
• Manual relation linking between notes
• Search with suggestions across notes, tasks, and decisions
• Qwen-powered note improvement for structure, type, status, tasks, and relations
• Human-reviewed memory patch apply flow: selected notes, tasks, and decisions become project memory
• Agent colors for Usuario, Claude, Codex, and Qwen
• Context Compiler with AGENTS.md and docs/context.md export
• SQLite models: Agent, Project, Note, Task, Decision, Relation, AgentRun

Not included:

• Cloud sync
• Mobile app
• Multi-user collaboration
• Canvas/graph editor
• Complex block editor

Setup

cd "D:\Neuronotes 2.0"
python -m venv .venv
.\.venv\Scripts\python -m pip install -r backend\requirements.txt
npm install
npm --prefix frontend install

Pull the default local model:

ollama pull qwen3:4b

Run In Browser

npm run dev

Backend: http://127.0.0.1:8787

Vite prints the frontend URL. It defaults to http://localhost:5173, but may use another port if 5173 is already busy.

Run As Desktop App

npm run desktop

This builds the React app, starts the local FastAPI backend, and opens Electron.

Model Settings

Default:

• Provider: ollama
• Base URL: http://localhost:11434
• Model: qwen3:4b

LM Studio/OpenAI-compatible mode:

• Provider: lmstudio
• Base URL: usually http://localhost:1234
• Model: the loaded model name shown by LM Studio

You can edit these from Settings in the app.

Useful Commands

npm run build
npm run mcp

Agent Access

Neuronotes exposes shared memory without binding it to one LLM vendor.

• MCP server: npm run mcp
• REST base: http://127.0.0.1:8787
• Search: GET /api/search?query=...
• Project notes: GET /api/notes?project_id=...
• Create note: POST /api/notes
• Update note: PUT /api/notes/{note_id}
• Improve note: POST /api/notes/{note_id}/improve
• Relations: GET/POST /api/relations
• Tasks: GET/POST/PUT /api/tasks
• Context handoff: POST /api/context/compile
• Proposed writeback: POST /api/memory-patches

Agents should search first, prefer approved/canonical items, keep source attribution, create or update durable notes, link related notes, and create tasks from actionable work.

Exported Context

Context Compiler writes files to:

exports/codex/AGENTS.md
exports/codex/docs/context.md
exports/codex/docs/decisions.md
exports/codex/docs/tasks.md