Slothly - The Overnight Agent That Works While You Sleep
Theme: Slow and Steady (International Sloth Day, Oct 20)
Most AI agents are built to move fast — a firehose of notifications, always-on cloud calls, per-token billing, data leaving your device. This week, I built the sloth edition.
What I built
Slothly is a fully local, overnight AI agent.
- You drop a voice note or written note into
inbox/before bed - Overnight, Slothly transcribes (with faster-whisper) and summarizes (with llama3.2 via Ollama) into a crisp two-line morning report
- It writes the report to
reports/YYYY-MM-DD.md, archives the source file, and even adds a calming sloth-themed haiku - Runs on a schedule via Windows Task Scheduler
The Stack (All Open Source)
- Agent harness: OpenAI Agents SDK (open source)
-
LLM inference: Ollama + open-weight model
llama3.2 - Transcription: faster-whisper (open source)
- Runtime: 100% local, CPU-only
Why open innovation matters
- Runs on a laptop with no internet — After the one-time model download, everything runs locally. No cloud APIs required.
- Keeps data off servers you don't control — Notes, voice memos, and summaries never leave your device.
- Fine-tune, swap models, or change agent behavior — Being able to swap open-weight models or tweak the agent logic is a first-class feature, not a locked-down API.
- Cost nothing to run — Local inference means zero API spend.
Where the open approach worked better than a closed one
A closed stack would have meant cloud transcription + cloud LLM + API keys + per-run cost + third-party data flow. With Slothly, the entire loop — transcribe, summarize, report — runs offline, on-device, for free. The open pieces made this simple lazy automation possible in a way that would be impractical (and expensive) with closed APIs.
The Sloth Philosophy
The agent moves slow and steady, conserves energy, and does the work so you don't have to. You sleep; the agent works. You wake to a two-line report — and maybe a sloth haiku.
Repo: https://github.com/MAXxDEVIL/Slothly
Built with open-source AI for Hacktoberfest Week 3. 🦥
Top comments (0)