Overnight Sloth: Wake Up to What Matters, Ignore the Noise

Overnight Sloth — Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady.

What I Built

I built Overnight Sloth, a local background agent that saves you from wasting mental energy on digital clutter. Every night, it ingests chaotic group chat exports, long email threads, or cluttered message logs, filters out the memes and small talk, and leaves a clean, two-bullet summary waiting on your desktop when you wake up.

Instead of starting your morning scrolling through 300+ unread messages just to see if plans changed or if you were assigned a task, Overnight Sloth does the heavy lifting while you sleep, rewarding your inner sloth with zero-effort morning catch-ups.

Code

GitHub Repository — Overnight Sloth

How I Built It

Overnight Sloth is built entirely around an open-source, local-first AI stack:

• LLM Engine: Powered by Ollama, running a quantized Llama-3.1-8B-Instruct model locally on-device.
• Agent Framework: Built using CrewAI to orchestrate a two-stage multi-agent pipeline:
  1. The Noise Filter Agent: Parses raw logs to strip out emojis, sticker spam, and irrelevant banter.
  2. The Executive Summarizer Agent: Condenses the core discussion into high-signal takeaways and action items.
• Automation: A background Python script using the watchdog library monitors local export directories overnight, automatically kicking off processing whenever new chat logs are dropped in.

Why Does Open Innovation Matter?

Open innovation was an absolute non-negotiable for this project for three critical reasons:

1. Data Sovereignty & Privacy: Group chats with family, friends, or close teams contain intimate conversations, phone numbers, and personal details. Sending raw personal chat transcripts to a closed third-party cloud API is a major privacy risk. With local inference, zero bytes of private conversation ever leave my machine.
2. Zero Running Costs: Running local quantized open-weight models means this automated morning routine runs every single day for free—no monthly API subscription fees, rate limits, or token costs hanging over my head.
3. Full Behavioral Control: Open-source agent frameworks allowed me to fine-tune system prompts and context handling directly, tweaking how aggressively the bot filters chat noise without hitting closed-system guardrail overrides or arbitrary token limits.

The Lazy Payoff

This week, I tested it on a high-activity group chat that accumulated over 400 messages overnight. Instead of spending 15 to 20 minutes reading through banter, I opened my desktop to a precise two-line brief highlighting a single deployment deadline change and one direct question addressed to me.

It instantly eliminated morning information overload and gave me back calm, uninterrupted focus to start the day.

Prize Categories

• Open-Source AI Challenge — Slow & Steady Theme
• Local-First & Privacy-Focused AI