A familiar chat
Answers stream in as they are written, with stop, copy, regenerate and edit-and-resend. Models that can “think” get a deep-thinking switch.
Lemon · local AI chat
AI chat on your own PC, offline
A ChatGPT-style Windows app for AI models that run on your computer through Ollama. One small .exe, in Arabic and English, with no account and no cloud.
Free & open source (MIT) · Windows 10/11, 64-bit · single file, no installer · needs Ollama
Lemon
Features
Everything stays on your PC
Lemon talks only to the Ollama server on your own computer. Your prompts, files and chats never leave it.
Assistants from your files
Give an assistant your PDFs, text, code or notes. It answers only from them in strict mode and shows its sources under each answer.
Live code preview
Run HTML, JavaScript or SVG, from pages to small games, in a sandboxed side panel. Errors are caught, with an “ask for a fix” button.
Models that fit your PC
Download and delete Ollama models inside the app. Suggestions are rated against your graphics card and memory: recommended, slow, or not recommended.
Arabic and English
Switch the language live in Settings, with correct right-to-left and left-to-right layout everywhere. Light, dark or system theme.
Private and up to date
No telemetry, accounts or ads. Its local server listens on 127.0.0.1 only and refuses requests from other websites. Updates install in one click, checked with SHA-256.
New in 2.2
Assistants that read your documents properly
With deep understanding, on by default for new assistants, the model reads each document once and writes a structured digest of its sections, numbers and conditions.
- Better answers to broad questions: follow-up questions are rewritten into full searches, and matching passages come with their neighbours.
- See what it read: the extracted text and the digest of each file can be viewed and edited, with a test panel for the assistant.
- Ready-made instructions: templates for legal, tutoring, support, research, HR and technical documentation.
Indexing takes longer with deep understanding: in the developer’s test on an RTX 2070 (8 GB) with qwen3:8b, a 6-page Arabic PDF took about 95 seconds. Scanned PDFs without a text layer can’t be read; use a copy with OCR text.
Lemon
Requirements
What you need
Lemon is the window you chat in. The AI models themselves run in Ollama, a free program that you install once.
Your PC
- Windows 10 or 11, 64-bit.
- Ollama installed on the same PC, from ollama.com. Lemon starts it for you if it is installed but not running.
- Microsoft Edge WebView2, already part of Windows 11 and up-to-date Windows 10.
- Recommended: a graphics card with 8 GB of memory or more and 16 GB of RAM. Small models also run on the processor alone, just slower.
Get started in three steps
- Install Ollama from ollama.com.
- Download Lemon and run
local-ai-chat.exe. There is no installer: keep the file wherever you like. - Open Models in Lemon and download one of the models suggested for your PC, for example a small Qwen or Gemma model.
Download Lemon
Free and open source. One file, straight from the app’s official release.
The release notes show the file’s SHA-256 fingerprint. To compare it, run Get-FileHash .\local-ai-chat.exe in PowerShell.
Does Lemon need the internet?
No. Chats run on your PC. The internet is used only when you download a model (Ollama fetches it) and for the update check, at start-up and every 6 hours, which you can turn off in Settings.
What is Ollama, and why do I need it?
Ollama is a separate, free program that runs AI models on your computer. Lemon is the app you chat in; it connects to Ollama at http://127.0.0.1:11434 by default, and you can change that address in Settings. Ollama and the models are not part of Lemon, and each model has its own license.
Which model should I start with?
Open Models in Lemon: the catalog rates each model against your graphics card and memory. A small Qwen or Gemma model is a good start. Models with 1 to 4 billion parameters also run without a graphics card, only more slowly.
Where are my chats stored?
In %APPDATA%\LocalAiChat on your PC: one file per conversation, plus settings and assistants. Exports go to Documents\LocalAiChat. To keep everything in one folder instead, set the LOCALAI_DATA_DIR environment variable.
Can an assistant read scanned PDFs?
Not yet. A scanned page is only an image, so there is no text to read. Give the assistant a copy of the PDF with OCR text and it works as usual.
Windows says “Windows protected your PC”. Is it safe?
Lemon is not code-signed yet, so SmartScreen doesn’t recognise the publisher. Click “More info”, then “Run anyway”. The code is public, every release is built automatically from it, and you can compare the file’s SHA-256 fingerprint with the one on the release notes.
How do I remove it?
Close Lemon and delete local-ai-chat.exe. To also delete your chats and settings, remove the LocalAiChat folders. Your models stay in Ollama until you delete them there or in Lemon’s Models page.