Scribe

Scribe is a privacy-first, desktop-based meeting minutes generator. Built with Wails (Go + React), it uses local AI via Ollama to instantly transform your messy meeting notes into structured, professional summaries without ever sending your data to the cloud.

---

Features

• 100% Local & Private: Your meeting notes never leave your computer. Everything runs locally on your own hardware.
• AI-Powered: Uses local LLMs (defaults to qwen3:8b) to intelligently format notes, identify action items, and flag things that need confirmation.
• Smart Glossary & Auto-Correction: Reads from custom local JSON files to enforce company-specific acronyms, vocabulary, and grammar rules.
• Keyboard-First Workflow:
  • Ctrl + Enter / Cmd + Enter to Generate
  • Ctrl + S / Cmd + S to Save as Markdown
  • Ctrl + C / Cmd + C to Smart-Copy Output
• Auto-Updates: Built-in checker notifies you when a new release is available on GitHub.

---

Project Background

Scribe originally started as a tool built at the request of a friend who needed a way to improve and standardize Indonesian-language text using a local AI model.

The project is developed free of charge as a personal project. At the same time, it serves as an opportunity for me to explore and learn more about local Large Language Models (LLMs), prompt engineering, AI-assisted text processing, and building applications around locally hosted AI.

---

Prerequisites

Scribe uses Ollama for its local AI processing.

Because the AI model runs locally on your own hardware, performance depends on your system's CPU, GPU, RAM, and available resources.

⚠️ Performance Note: Scribe currently uses Qwen3 8B, which is a relatively large local model. Lower-end laptops and desktops may experience slower response times, longer processing times, or higher system resource usage. Systems with more RAM and a capable GPU will generally provide a smoother experience.

1. Download and install Ollama.

2. Open your terminal and download the model used by Scribe:

   ollama pull qwen3:8b

3. Verify that the model is available:

   ollama list

   Make sure qwen3:8b appears in the list before running Scribe.

💡 For lower-end hardware: You can try Qwen3 4B Instruct if Qwen3 8B runs too slowly on your system. Scribe has not been extensively tested with other models yet, so compatibility and output quality may vary.

---

Installation

You don't need to install any additional dependencies to use Scribe.

1. Go to the latest release.
2. Download the appropriate archive for your operating system:
   • Scribe-Windows.zip — Windows
   • Scribe-macOS.zip — macOS
3. Extract the downloaded archive.
4. Run the Scribe application.

macOS: Opening Scribe

⚠️ One-time setup: The following steps only need to be completed once. After completing them, you can open Scribe normally without repeating these steps.

Because Scribe is currently distributed without Apple Developer ID notarization, macOS may prevent the application from opening normally.

If macOS blocks Scribe, follow these steps to allow the application to run.

1. Open the app normally so the error pops up, then click Done or Cancel
2. Click the Apple menu in the top left corner and select System Settings.
3. Click on Privacy & Security in the left sidebar.
4. Scroll down to the security section. You will see a message saying the app was blocked from use.
5. Click the Open Anyway button next to it.

Once completed, you should be able to launch Scribe normally from Finder or your Applications folder.

Note: This warning is caused by macOS security protections for applications that have not been notarized by Apple. It does not necessarily indicate that there is a problem with Scribe.

---

Customizing AI Behavior

Scribe is currently designed primarily for Indonesian-language text and provides built-in tools for customizing how the AI processes and generates text.

Glossary

Add custom terminology, abbreviations, and acronyms that are specific to your organization or project.

This helps Scribe understand terms that may not be part of the AI model's default vocabulary and maintain consistent terminology throughout its output.

Normalization

Define common variations, slang, or informal Indonesian expressions and their preferred forms.

This allows Scribe to normalize inconsistent input before it is processed by the AI, helping produce cleaner and more consistent results.

Both features can be managed directly through Scribe's interface without modifying configuration files or source code.

Failures

Provide examples of common mistakes that the AI should avoid.

You can define a wrong example alongside the correct version, allowing Scribe to use these examples when generating or correcting text.

---

Privacy

• No tracking
• No third-party AI API calls
• No data sent to OpenAI, Anthropic, or other external AI providers

Scribe's AI processing is performed using a locally hosted AI model, keeping the text processing on the user's machine.

The only external network request made by Scribe is an anonymous request to GitHub to check whether a newer version of the application is available.

No user-generated content is transmitted as part of this update check.