Slow and Steady: Building a Local AI Group Chat & Task Summarizer

What I Built

I created a lightweight local automation concept that processes long overnight group chat conversations and messy unread messages. Instead of waking up to hundreds of unread messages, this tool generates a quick 2-line summary along with an actionable to-do list for the day.

Key Features:

• Morning Briefing: Automatically summarizes overnight messages into a concise 2-sentence snapshot.
• Action Extraction: Identifies pending tasks, deadlines, and direct mentions so nothing gets missed.
• Offline Processing: Operates directly on my local machine using open-weight models.

How It Works

1. Input: Takes exported group chat logs or text notifications from overnight chats.
2. Local Inference: Passes the text through a small, quantized open-weight model (like Llama 3 or Mistral) running locally via Ollama.
3. Output: Formats the output into a clean, sloth-friendly morning summary.

Why Open Innovation & Open-Weight Models Matter

Using an open-source, local-first approach made a huge difference compared to relying on proprietary cloud APIs:

1. Absolute Privacy: Group chat logs contain personal conversations, contact details, and private banter. Running an open-weight model locally ensures private messages never leave my laptop or get sent to third-party servers.
2. Zero Operating Cost: Proprietary AI APIs charge per token. A local model costs nothing extra to run after setup—build once, relax forever.
3. Offline Reliability: It processes data seamlessly without needing a high-speed internet connection, allowing the automation to run steadily at any time.