notesci

notesci is an open-source, local-first research workspace for reading, organizing, and writing with your own scientific materials.

It combines a desktop app, a FastAPI backend, Postgres/pgvector retrieval, multi-provider chat models, local file ingestion, citations, MCP tools, and a curated marketplace for skills, plugins, and MCP servers.

Download

Installers are published on GitHub Releases:

https://github.com/psypeal/notesci/releases/tag/v0.1.0

Current packages:

Platform	Package
Ubuntu / Debian	.deb
Fedora / RPM Linux	.rpm
macOS Apple Silicon	.dmg
Windows x64	.exe NSIS installer

The Linux AppImage build is still treated as best-effort. Use the .deb or .rpm packages for local testing today.

What notesci does

• Ingest PDFs, text, web pages, and project materials into a private research library.
• Chat with your uploaded sources using retrieval-grounded answers and citations.
• Browse internal source citations and external web sources from the chat UI.
• Use multiple model providers: Anthropic, OpenAI, Google, and DeepSeek.
• Configure custom chat and embedding providers where supported.
• Install curated MCP servers, plugins, and skills through a marketplace-style interface.
• Use local MCP integrations such as Zotero and Obsidian when configured.
• Draft and polish long-form research writing with a multi-stage workflow.

Project status

notesci is early-stage open-source software. The v0.1.0 packages are intended for local testing and feedback, not production-critical research workflows.

Known rough edges:

• Some backend tests and lint checks are currently advisory while older invite-era tests are updated.
• Code signing is not configured yet, so macOS and Windows may show unsigned app warnings.
• AppImage packaging is not the primary Linux distribution path yet.

Architecture

Area	Stack
Desktop shell	Tauri
Frontend	React, TypeScript, Vite
Backend	Python, FastAPI
Agent runtime	LangGraph, LangChain
Storage	Postgres, pgvector
Retrieval	chunked project materials + configurable embedding model
Integrations	MCP servers, local plugin/skill marketplace

The desktop packages bundle the app shell and required runtime resources. The Linux .deb package uses system Postgres dependencies; cross-platform bundles embed more runtime components for one-click local testing.

Repository layout

Path	Purpose
backend/	FastAPI API, migrations, retrieval, chat agent, MCP host
frontend/	React workspace, chat UI, reader, marketplace, settings
desktop/src-tauri/	Tauri desktop app and installer configuration
packaging/	Linux/macOS/Windows packaging helpers
landing/	Static project landing page
notesci/design_handoff_notesci_logo/	Canonical logo, wordmark, and brand tokens
.github/workflows/	CI and release packaging workflows

More detailed implementation notes live in:

• backend/README.md
• frontend/README.md
• packaging/cross-platform.md
• landing/README.md

Brand assets

The canonical notesci mark is the teal + indigo NS lettermark with the lowercase notesci wordmark. Source SVGs live in:

notesci/design_handoff_notesci_logo/assets/

Recommended assets:

Asset	Use
lockup-teal-indigo.svg	README, landing page, headers
icon-teal-indigo.svg	Default app icon / favicon
icon-app-tile.svg	Dark app tile and store-style contexts
icon-mono.svg	Single-color print or mask contexts

See notesci/design_handoff_notesci_logo/README.md for geometry, colors, typography, and implementation guidance.

Local development

Backend:

cd backend
uv sync
cp .env.example .env
uv run fastapi dev src/notesci/main.py

Frontend:

cd frontend
pnpm install
pnpm dev

Desktop:

cd desktop/src-tauri
cargo tauri dev

For local retrieval and ingestion, configure Postgres with pgvector and set the required provider keys in the backend environment.

Release process

Release packages are built by .github/workflows/release.yml.

The current public release is:

v0.1.0

The workflow builds:

• Linux .deb and .rpm
• macOS .dmg
• Windows .exe

Tagging v* starts a release build and attaches produced artifacts to the matching GitHub Release.

License

notesci is open source under the Apache License 2.0. See LICENSE for the full license text.