Vimal Sahani

Posted on Oct 21

SlothPilot: The Autonomous Overnight Multi-Agent Command Center (Because Developers Deserve 8 Hours of Sleep)

#devchallenge #hf26challenge #python #ai

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

This is a submission for the Hacktoberfest 2026 Week 3 DEV Challenge: Slow and Steady

"Channel your inner sloth. You sleep 8 hours, SlothPilot works overnight."

In celebration of International Sloth Day (October 20) and the Hacktoberfest Week 3 Challenge ("Slow and Steady — Build something with open-source AI that rewards laziness"), I built SlothPilot.

🌐 Live Interactive Demo & Open-Source Code

Experience the Morning Command Center live in your browser:

👉 Live Streamlit App: https://slothpilot-xaqtgwwbpv2tfvscjhxakr.streamlit.app/
👉 GitHub Repository: https://github.com/VimalN2005/slothpilot

(You can simulate an overnight run right from the sidebar, inspect raw ingested sources, test the 1-click Approval Gate, generate an instant morning standup, listen to the 45-sec voice briefing, and inspect live OpenTelemetry trace spans!)

🥱 The Problem: The 9:00 AM "Digital Archaeology" Nightmare

Every software engineer knows the dreadful morning routine:
You sit at your desk at 9:00 AM with a cup of coffee. Instead of diving into creative, high-impact architecture, you spend your first 60 to 90 minutes performing stressful digital archaeology:

Dirty Git Branches: Which feature branches were left uncommitted before shutting the laptop last night?
Chaotic Chat Backlogs: What did teammates, managers, or clients message on WhatsApp, Discord, or Slack overnight? Did someone ping about an urgent payment webhook failure or an impending contract milestone?
Fragmented Notes & Todos: Scattered Markdown files, half-written bullet points in Obsidian/Notion, and priority tags (!high, p1, urgent) spread across multiple directories.
Calendar Anxiety: When is my first meeting today, and how much time do I actually have before I need to be camera-ready?
Standup Fatigue: Typing your morning standup update from memory while half-asleep.

By the time you finish triaging this chaos, your morning energy is drained. You've answered messages hastily, context-switched five times, and haven't written a single line of thoughtful code.

⚖️ The Contrast: The 9:00 AM Grinder vs The SlothPilot Developer

Feature	The 9:00 AM Corporate Grinder 😫	The SlothPilot Developer 🦥☕
Wakeup Ritual	6:30 AM cold panic checking Slack	9:45 AM (Enjoying +120m Sleep Buffer)
Git Archaeology	40 mins manually checking dirty branches	1-click approved auto-checkpoint
Client Triage	Scrambling 15 browser tabs & chat exports	Pre-drafted contextual replies ready
Daily Standup	15 mins typing updates from memory	1-click auto-generated Standup copied
Morning Briefing	Reading dense terminal diffs	🎙️ 45-sec calm audio briefing while brushing teeth
Privacy & Cost	\$50/mo cloud API + code sent to 3rd party	\$0.00 USD + 100% Local On-Device

💡 What I Built: SlothPilot 🦥

SlothPilot is an autonomous overnight multi-agent briefing system built with open-source AI and Python.

Before going to bed, you either let the automated scheduler trigger it at 03:00 AM or run a single command. While you sleep 8 uninterrupted hours, SlothPilot quietly inspects your local workspace, synthesizes overnight developments across 5 data sources, pre-drafts contextual replies and git actions, calculates your Sloth Sleep Buffer, and stages everything in a clean Morning Command Center with a strict 1-click Human Approval Gate.

When you wake up, you don't scramble through 10 browser tabs. You open your Morning Dashboard, listen to your audio briefing, copy your auto-standup, review your Top 3 Daily Focus Items, click Approve on pre-drafted client emails, and save 2+ hours every morning.

🏗️ How It Works: The Multi-Agent Pipeline

SlothPilot does not rely on a single massive prompt. Instead, it coordinates three specialized local agents orchestrated in a deterministic pipeline:

Local Data Sources (Git Repos, Notes, Chats, Todos, Calendar .ics)
        │
        ▼
   Ingestion Layer (app/ingestion)
   ├── Git Scanner (subprocess diffs & status)
   ├── Markdown Vault Parser (tasks & action items)
   ├── Chat Export Parser (WhatsApp/Discord urgent patterns)
   └── Backlog Parser (priority tags & deadlocks)
   └── Calendar Parser (.ics meeting timeline & Sleep Buffer)
        │
        ▼
   Multi-Agent Pipeline (app/agents)
   ├── 🧠 Context Aggregator Agent
   ├── ✍️ Action Proposer Agent
   └── 🛡️ Safety & Priority Reviewer
        │
        ▼
   Persistent State Store (SQLite)
   └── Full OpenTelemetry GenAI Span Traces
        │
        ▼
   Morning Command Center (Streamlit Dashboard)
   ├── 🎙️ 45-Sec Sloth Audio Briefing Player
   ├── 📋 1-Click Auto-Standup Generator (Slack / Discord)
   ├── 🛌 Sloth Sleep Buffer & Meeting Timeline
   ├── 🛡️ Human Approval Gate (1-click Approve / Dismiss)
   ├── 🔍 Ingested Workspace Inspector (5 Source Tabs)
   └── 📊 OpenTelemetry Observability (Latency & Zero-Cost Metrics)

1. The Context Aggregator Agent (`app/agents/aggregator.py`)

Scans raw local artifacts and extracts structured signals:

Uncommitted git diffs, dirty working trees, and recent commit messages.
High-priority chat patterns (BUG, URGENT, CONTRACT, PAYMENT, DEMO).
Open - [ ] markdown checkboxes and priority backlog todos (!high).
Upcoming morning meetings and schedule constraints from .ics exports.

2. The Action Proposer Agent (`app/agents/action_proposer.py`)

Rather than merely summarizing, this agent pre-drafts concrete solutions:

Client & Team Replies: Contextual, empathetic message drafts acknowledging the issue with specific next steps (e.g. SOW advance wire confirmation, payment webhook triage).
Git Workflows: Stash and branch commands (git add -A && git commit -m "checkpoint: overnight triage") to keep working trees clean.
Task Scheduling: Converts urgent messages directly into prioritized daily action cards.

3. The Safety & Approval Reviewer Agent (`app/agents/reviewer.py`)

Crucially, SlothPilot never executes side-effects unprompted. The reviewer agent:

Assigns safety ratings (SAFE, REQUIRES_REVIEW, DANGEROUS).
Calculates the Sloth Efficiency Score (measuring developer energy conserved).
Calculates the Sloth Sleep Buffer: If your first meeting isn't until 10:30 AM, SlothPilot informs you that you have +120 mins of extra guilt-free sleep!
Enforces the Human Approval Gate: all proposed actions remain staged until you explicitly click Approve or Dismiss.

👑 The Two Ultimate Laziness Features:

1. 📋 1-Click Auto-Standup Generator

Never rack your brain trying to remember what you committed yesterday or what meetings are scheduled today. SlothPilot generates an executive bulleted standup ready to paste directly into Slack or Discord:

🦥 Morning Standup Update (via SlothPilot):

🚀 Yesterday / Overnight:
- Triaged uncommitted git changes and overnight chat backlogs
- Automated test suite verification across workspace

🎯 Today's Focus:
- Review payment webhook queue blocker before 10 AM demo
- Approve and dispatch SOW contract reply to Client Dave
- Attend: Sprint Architecture & Payment Webhook Review (10:30 AM)

🛡️ Blockers:
- Needs approval: 2 pre-drafted client/team communications

2. 🎙️ 45-Second Voice Morning Briefing

For developers who refuse to look at screens before their first coffee, the dashboard includes a native browser voice player that reads your efficiency score, upcoming meeting time, and staged action counts aloud.

🛡️ The Human Approval Gate: Zero Autonomous Disasters

A major danger of autonomous AI agents is unprompted side-effects—an agent accidentally sending a premature email to a client or pushing broken code to production.

SlothPilot solves this with a Zero Unprompted Side-Effects Guarantee:

Every action card displays its category (COMMUNICATION, GIT_ACTION, TASK_RESOLUTION), safety rating, reasoning, and target destination.
With one click in the Streamlit UI, you can approve or dismiss each item.
Your morning workflow becomes: Review -> Approve -> Relax.

⏰ Set & Forget: Automated Overnight Scheduler Daemon

Prefer never to type a command before sleeping? SlothPilot includes an automated scheduler daemon (scripts/schedule_overnight.py):

python scripts/schedule_overnight.py --hour 3 --minute 0

SlothPilot quietly wakes up at 03:00 AM every night, runs the local ingestion and multi-agent pipeline, and stages your Morning Briefing before your alarm rings.

📊 OpenTelemetry Observability: Tracking Latency & Zero Costs

SlothPilot features built-in OpenTelemetry instrumentation (app/core/telemetry.py):

Every pipeline phase is tracked with W3C-compliant trace and span IDs (slothpilot.overnight_pipeline, slothpilot.ingestion, slothpilot.propose_actions, slothpilot.reviewer).
Real-time metric comparison: displays $0.00 USD inference cost vs the proprietary API tax you would have paid on closed cloud models.

🔓 Why Open Innovation & Open-Source AI Matter

SlothPilot operates on the most sensitive assets a developer possesses: uncommitted source code, private team chat exports, calendar meetings, and personal notes.

Sending this private data to cloud-hosted proprietary APIs poses significant privacy risks and unpredictable subscription fees. By utilizing open-weight models (via Ollama / Gemma / Llama):

100% Privacy & Zero Data Exfiltration: Your code, client conversations, and notes never leave your local machine.
$0.00 API Cost: Run overnight autonomous pipelines every single night without paying a dime in API bills.
Offline Autonomy: Works reliably even on flights or when your local network is down.

💻 Tech Stack

Language & Runtime: Python 3.10+
Data Validation & Schemas: Pydantic v2
Dashboard UI: Streamlit
Voice Synthesis: Native Web Speech API
State Persistence: SQLite
Observability: OpenTelemetry GenAI Spans
AI Core: Ollama / Open-Weight Models (Llama 3.1, Gemma 2)
CI/CD: GitHub Actions Matrix Testing (Python 3.10, 3.11, 3.12)
Test Suite: Pytest (13/13 unit tests passing in ~1.0s)

🦥 Final Thoughts: Embrace the Sloth Mindset

True engineering productivity isn't about waking up at 5:00 AM to answer 200 unread messages in a panic. It is about building resilient open-source tools that work diligently in the background so you can enjoy your sleep, wake up with clarity, and focus on what truly matters.

Happy Hacktoberfest 2026, and happy International Sloth Day! 🦥☕

Top comments (2)

Shipmark • Oct 21

Congrats on shipping SlothPilot. An approval gate before an overnight agent acts is the part most agent demos skip, and yours puts it front and center. Free offer from our side, no strings. Run your demo site through the quick check at shipmark.co. The one thing we'd look at first is your page source, checking that no API key or config sits in public view, since a tool that runs agents overnight gets judged on its own setup first. Free to run.

Vimal Sahani • Oct 21

Thanks so much! Really glad you liked the Human Approval Gate — keeping full developer control was priority #1.
The best part about SlothPilot is that it runs 100% locally with open-weight models (via Ollama), so there are zero cloud API keys or secrets to leak in the frontend!
Appreciate you checking it out!