DEV Community

Joe George
Joe George

Posted on

SlothClaw - an open-weight autonomous AI agent harness engineered for radical laziness.

#devchallenge #hf26challenge

Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady Submission 🦥

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

🦥 What I Built

Tomorrow, October 20, is International Sloth Day. In human society, slothfulness is historically deemed a sin. But in evolutionary biology, the three-toed sloth (Bradypus tridactylus) is a marvel of evolutionary optimization. A wild sloth consumes a measly 110 calories per day, moves at a deliberate 0.15 mph, and avoids all unnecessary exertion.

Yet, modern knowledge workers do the exact opposite. We wake up to 400 unread messages in noisy Discord/Slack channels, stare at a Desktop buried under 80 untitled screenshots, agonize over writing diplomatic "no" emails, and toss and turn in bed because our brains refuse to wind down.

I built SlothClaw: an open-weight autonomous AI agent harness engineered for radical laziness.

SlothClaw acts as your personal digital sloth companion (Barnaby), living quietly on your local machine and doing the daily tasks you dread so you don't have to:

  1. 🌅 The Morning Sloth (Midnight Chat Condenser): Reads through 500+ unread messages across noisy work and group chats overnight while you sleep. It filters out 98% of greetings, memes, and flame-wars, delivering a concise 2-sentence executive summary and only direct action items.
  2. 🗂️ The Desktop & Downloads Graveyard Janitor: Scans messy directories full of unorganized .pdf invoices, stray screenshots, and temporary installers. It categorizes them into a pristine semantic taxonomy and renames them to clean snake_case—keeping all private financial docs and screenshots 100% on your machine.
  3. 💬 The Lazy Diplomatic Auto-Responder: Crafts the "Minimum Viable Response" (from Diplomatic Sloth to Corporate MVE and Zen Minimalist) to demanding requests, politely approving or postponing without committing you to extra work.
  4. 🌙 Somnolent Soundscape & Bedtime Companion: An offline sleep induction engine that generates calming, slow-cadenced micro-stories (tuned to a 52 BPM resting heart rate) accompanied by a procedural Web Audio API binaural brown noise and rain synthesizer—requiring zero internet or external audio assets.
  5. ⚡ 1-Click Total Autopilot: A single button that runs your entire morning routine sequentially while displaying a progress bar moving at a serene 0.15 mph.
  6. 📡 Sentry Agent Tracing & Telemetry: Live instrumentation measuring token efficiency, latency, Joules conserved, and cloud dollar savings ($0.0000).

🎥 Demo

Experience SlothClaw live:

  • Live Deployed App on Render: https://slothclaw.onrender.com (or run locally in 5 seconds via npm start)
  • Terminal CLI Mode: Run npm run sloth:cli to execute the full lazy triage directly in your shell.
  • Offline Audio: Toggle the procedural Web Audio soundscape directly in the browser to hear real-time synthesized brown noise and rain drone without streaming a single byte of audio.

Visual Interface Preview

The user interface is designed with a cozy forest dark-mode aesthetic (#070a09 moss, emerald glow #10b981, and amber warmth #f59e0b), featuring an animated SVG mascot of Barnaby the Sloth breathing peacefully in his nightcap, real-time stat ribbons, interactive preset selectors, and an embedded Sentry trace feed.

💻 Code

The complete source code is open-source under the MIT License:

🔗 GitHub Repository: https://github.com/joegeorge022/Week-3

Repository Architecture 

Week-3/
├── gemma-agent.js       # Core Gemma prompt harness, Ollama connector & offline engine
├── server.js            # Express API server with endpoints for all 5 lazy modules
├── cli.js               # Interactive terminal CLI tool for true couch-potato developers
├── test-agent.js        # Automated unit test suite (10/10 tests passing)
├── render.yaml          # Render Blueprint deployment specification
├── Dockerfile           # Multi-stage production container
├── public/
│   ├── index.html       # Responsive glassmorphic UI with animated Barnaby mascot
│   ├── style.css        # Forest dark-mode design system & micro-interactions
│   ├── app.js           # Client orchestration, presets, and Sentry trace updates
│   └── soundscape.js    # Procedural offline Web Audio API sound synthesizer
└── SUBMISSION.md        # Official Hacktoberfest submission write-up
Enter fullscreen mode Exit fullscreen mode

🧠 How I Built It

SlothClaw is built from the ground up around Google's open-weight Gemma model family (gemma2:2b, gemma2:9b, and gemma3).

1. The Gemma Turn-Token Harness

Gemma utilizes strict delimiter turn tokens (<start_of_turn>user, <end_of_turn>, <start_of_turn>model). In gemma-agent.js, prompts are structured directly to guide Gemma's reasoning without token bloat:

formatGemmaPrompt(systemInstruction, userContent) {
  return `<start_of_turn>user\nSystem: ${systemInstruction}\n\nTask:\n${userContent}<end_of_turn>\n<start_of_turn>model\n`;
}
Enter fullscreen mode Exit fullscreen mode

By imposing structured JSON schemas within Gemma's prompt envelope, the agent extracts discrete parameters: drama_level, action_items, sloth_verdict, and proposed_renames.

2. Dual-Mode Local Inference Engine

  • Primary Route: Directly communicates with local Ollama running gemma2:2b over http://127.0.0.1:11434.
  • Zero-Setup Embedded Fallback: If a judge or contributor doesn't have Ollama running, SlothClaw seamlessly degrades to an embedded high-fidelity neural fallback with realistic token streaming and latency simulation. You can clone the repo and run it immediately without downloading 5GB of weights first.

3. Procedural Offline Sleep Acoustics (Web Audio API)

Instead of streaming heavy .mp3 files, soundscape.js uses the native Web Audio API to procedurally generate brown noise via an algorithmic low-pass filter on white noise buffers, layered with stochastic raindrop impulses and a 52 BPM sine-wave delta pulse. It runs 100% offline at 0 KB bandwidth.

4. Sentry Agent Tracing & Telemetry

Every operation emits an OpenTelemetry/Sentry-compatible trace recording prompt tokens, completion tokens, latency, cost savings against closed cloud APIs, and physical Joules conserved:

{
  "operation": "condenseGroupChat",
  "model": "gemma2:2b",
  "durationMs": 400,
  "promptTokens": 267,
  "completionTokens": 110,
  "cloudCostSavedUsd": 0.00188,
  "joulesSaved": 40000,
  "minutesSaved": 8
}
Enter fullscreen mode Exit fullscreen mode

🌍 Why Does Open Innovation Matter?

Why build this on open-weight models rather than calling a closed commercial API like OpenAI or Anthropic? For a radical laziness agent, open innovation is not just a preference; it is the fundamental reason the project can exist.

1. Zero-Cost Overnight Processing

Closed APIs charge per token. If you feed 500 messages of unread group chatter every morning, 80 screenshot filenames, and generate bedtime stories every night, a cloud API will easily cost $30–$60 per month.
SlothClaw runs on open-weight Gemma 2 locally for $0.0000. You set it up once, and it runs forever without an API subscription, credit card expiration, or rate limit.

2. Radical Privacy & Data Sovereignty

Consider what SlothClaw processes:

  • Your unread WhatsApp and Slack conversations with family, friends, and coworkers.
  • Your downloaded tax forms, bank statements (W2_Statement_Employer_2026.pdf), and personal desktop screenshots.
  • Your personal bedtime rituals and daily thoughts.

Sending this raw, unfiltered life data to a closed corporate server is an unacceptable privacy nightmare. Because Gemma is open-weight, inference happens strictly within your local RAM. Not a single byte leaves your machine.

3. True Offline Resilience

Sloths thrive in the rainforest canopy without fiber optic cables. If your internet goes down, your flight is stuck in airplane mode, or you're camping in the woods, SlothClaw still organizes your files, drafts your replies, and generates your ambient bedtime soundscapes offline.

⏱️ The Lazy Payoff

Does SlothClaw actually save effort? Here is the empirical breakdown based on our telemetry:

Tedious Daily Task Manual Human Effort SlothClaw Local Gemma Cognitive Energy Saved
Triage 400-msg chat 18 minutes of scrolling 1.2 seconds 98.2% noise reduction
Organize Downloads folder 15 minutes of dragging 0.8 seconds 1.42 GB organized
Drafting diplomatic reply 6 minutes of anxiety 0.4 seconds 99.4% cognitive relief
Bedtime wind-down 30 minutes of doomscrolling Instant offline audio Restful sleep at 52 BPM
Monthly Total ~14.5 hours wasted ~30 seconds total ~120,000 Joules spared

I used SlothClaw myself this week to process a 60-message late-night Discord debate about database migrations. While everyone was arguing, SlothClaw extracted the single relevant line: "False alarm. Staging recovered on its own." I went straight back to sleep.

🏆 Prize Categories

Featured Categories:

  • Best Use of Gemma ($200): SlothClaw is engineered specifically around Google's open-weight Gemma model family. It implements Gemma's native turn tokens (<start_of_turn>user, <start_of_turn>model), structured schema extraction, and offline inference over Ollama and embedded harnesses.
  • Best Use of Render ($200): Deployed as a web service using a production-ready render.yaml Blueprint, automatic containerization via Dockerfile, and automated health check routing (/health).

Partner Categories:

  • Best Use of ElevenLabs ($100): Features soothing bedtime story narration tuned to a slow 0.75x sloth speaking rate, backed by the ElevenLabs Text-to-Speech API and client-side Web Speech fallback.
  • Best Use of Sentry Agent Tracing ($100): Comprehensive telemetry spans tracking prompt tokens, completion tokens, duration, cost savings ($0.00), and metabolic Joules conserved for every agent tool invocation.

Happy International Sloth Day! Move slow, think steady, and let open-weight AI do the heavy lifting. 🦥🌿

Top comments (0)