NoteLah's original code is available under the MIT License. Third-party components retain their own licences. See third-party notices, privacy and storage, AI and usage limitations, and contribution guidelines. The software is provided as-is; keep independent backups and verify AI answers against their sources.
Maintainers: follow the public release checklist before publishing source or executable releases.
It is a personal project intended to solve my daily problem, usually i have tons to do and i didn't jot it down. Thus, i forgot. So this app helps track my task and give some sense of accomplishment on the profile page when we have the box with green and greener thing. similar like token used haha.
I also built NoteLah to learn retrieval-augmented generation (RAG) through a working application. Its local Chat implements document extraction, section-aware chunking, embeddings, hybrid retrieval using BM25 and vector similarity, reciprocal rank fusion (RRF), and answers with source references through Ollama. The project lets me explore retrieval quality, evaluate results, and understand the limitations of small local models. Implementation details and learning notes are in docs/RAG.md.
The desktop Chat also includes a bounded agent: it can inspect unfinished tasks, search notes, propose a day plan, or break a goal into suggested new tasks. Task additions and date changes require explicit approval. See agent implementation and safeguards.
Open outputs/NoteLah-0.1.0.exe. A build copy also lives in release/. This is a portable Windows application; Node.js and a web server are not needed to use it. The first launch opens a blank workspace. The optional sample workspace is a temporary demonstration and never changes your own saved data.
To pin NoteLah to the Windows taskbar, keep the portable .exe in a permanent folder, open it, then right-click its taskbar icon and choose Pin to taskbar. If you pinned an older build and Windows says the executable is missing, unpin that shortcut and pin again after opening the updated build. Moving or renaming the portable .exe also requires pinning again.
- Today: add tasks, keep notes, choose dates, complete work, and bring unfinished tasks forward.
- Later: capture ideas without choosing a date, then schedule them when ready.
- Notes: create folders and written notes, import
.txt,.md, and PDF files, search the library, and move notes between folders. Text imports have a read-only text preview; PDFs open in NoteLah’s built-in viewer. Chat extracts PDF text with page references for local AI search. - My profile: edit your name and bio, see completed work, and click a day in the green activity grid to revisit it.
- Settings: choose the start of the week, manage backups, restore, and export.
Written notes and folder information are stored in %APPDATA%\NoteLah\notelah.sqlite. Imported originals are copied into %APPDATA%\NoteLah\files using content-based filenames, so moving your source files does not break the library. Settings shows the exact folder and imported-file usage. The first upgrade saves a recovery copy of the old database before adding Notes.
Use Create backup or Export JSON to make a complete copy, including imported files. Copying only the live database file does not include those external copies. Old task-only backups still restore. Deleting a folder moves its notes into Inbox; deleting a note removes its managed copy when nothing else uses it, while your original source file and existing complete backups are kept.
Import limits: UTF-8 .txt/.md up to 1 MB each, PDFs up to 20 MB each, 10 files per import, and 50 MB of distinct imported files in the library. Backups are limited to 100 MB. The browser version supports written notes and folders; importing documents and restoring backups containing them requires the Windows app.
Requires Node.js 22.12 or newer and npm.
npm ci
npm run devOpen the local URL printed by Next.js. All fonts are bundled locally.
npm run typecheck
npm test
npm run build
npm run preview
npm run desktop
npm run package:windesktop builds the Next.js static export and opens it in Electron. package:win builds the portable Windows executable in release/. Development and packaging require downloading open-source dependencies; the packaged application runs offline.
The desktop app is intended for one person on one laptop. Chat implements local hybrid RAG (BM25 + vector similarity + reciprocal rank fusion) and a day-planning agent using Ollama. Cloud synchronization is not implemented. The Windows build is unsigned.
Open-source acknowledgments are in THIRD_PARTY_NOTICES.md. Dependency versions are locked in package-lock.json.
Choose Ask my notes for document questions or Plan my day to inspect unfinished tasks and propose a plan. The planner takes a date, available minutes, and an optional session preference. Review its task choices, estimated durations, date changes, activity log, and consulted notes before clicking Apply plan. Applying saves only task planned dates; durations and order stay in the chat. A complete backup precedes each apply/undo. Undo is available in the current session only while the workspace has no newer edits. See agent details and limitations.
Choose Break down a goal, or ask "what tasks do I need?" in Plan my day, for a suggested checklist. It can propose general research actions without a document, with assumptions and missing requirements labelled for verification. Add tasks saves the reviewed checklist in Later, without scheduling it or changing existing tasks. The whole checklist is not constrained to a single day's time budget. No web browsing is performed.
Open Chat in the Windows app. Choose an installed chat model and embedding model, select all notes or a folder, and ask a question. The first question builds a local search index; later questions reuse unchanged embeddings. Ollama must be running. The initial setup uses qwen3.5:2b-q4_K_M for answers and nomic-embed-text:latest for embeddings (ollama pull nomic-embed-text). Models are downloaded once; inference uses only 127.0.0.1:11434. Cloud models are excluded.
Answers include clickable source labels, section-aware passages, BM25/vector ranks, RRF scores, and expandable search details. Independent questions exclude earlier answers; referential follow-ups resolve into standalone searches. Structured claims require source numbers, and inferred conclusions are labelled. Stop cancels the current job. Chat history lasts only for the app session. The derived %APPDATA%\NoteLah\search-index.sqlite is separate from the notes database and is rebuilt after restoration as needed; it is not included in backups. Scanned PDFs need OCR, which is not included. See RAG implementation and learning notes.