OutOfOffice AI: Step Away from Your Desk with Local Gemma 2 and Agent Telemetry

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

Most modern AI coding assistants do the exact opposite of encouraging you to touch grass: they demand your constant attention. You write a prompt, watch tokens stream, babysit diffs line-by-line, and wait on progress spinners.
OutOfOffice AI (OOO) was engineered to break this screen dependency.
It is an autonomous, local-first developer agent powered by Google Gemma 2. Instead of micromanaging an AI in an IDE window, you:

1. Target any local Git repository and set a maintenance directive (e.g., prune dead code, fix broken imports, run test suites).
2. Close your laptop, step away from your desk, and touch grass.
3. Return to an executive ElevenLabs voice debrief summarizing codebase health and modified files, backed by verified Git patches isolated in dedicated branches. By turning asynchronous coding chores over to an autonomous local agent, your screen time becomes the shortest part of the development loop.

---

Demo

Youtube Video:

What You See in the Demo:

• OOO Landing Page: Clean, developer-focused overview showing the local-first architecture and guarantees.
• Mission Control: Local Git manifest intake and zero-data-egress model selection.
• Sentry Telemetry Waterfall: Microsecond span instrumentation across PlannerNode, ASTScanNode, and GitBranchIsolation.
• ElevenLabs Voice Debrief: Executive audio briefing summarizing repository health score and tasks completed.
• Unified Diff Viewer: Syntax-highlighted git diffs isolated in the ooo/agent-audit branch.

---

Code

GitHub Repository:

Samar-365 / OutOfOffice-

Local Autonomous Coding Agent

OutOfOffice AI (OOO)

Local-First Autonomous Coding Agent Powered by Google Gemma 2, Sentry Agent Tracing, and ElevenLabs.

Overview

Modern AI coding tools suffer from a common limitation: they demand constant developer presence. Developers spend hours waiting for tokens to stream, reviewing single-line completions, and babysitting chat prompts.

OutOfOffice AI (OOO) flips this workflow. It is an autonomous, local-first developer agent designed for asynchronous execution:

1. Target any local Git repository and set a maintenance directive (dead code pruning, AST refactoring, test repair).
2. Step away from your desk while the agent executes locally on your hardware.
3. Return to an executive ElevenLabs audio debrief and inspect verified, branch-isolated syntax diffs.

No proprietary code is transmitted to cloud LLM APIs. All model inference runs on your machine via Google Gemma 2.

---

Core Guarantees

• 100% Local Intelligence: Runs Google Gemma 2 locally through Ollama. Zero cloud token costs, zero latency overhead, and complete data privacy.
• …

View on GitHub

The codebase is structured into:

• backend/: FastAPI daemon, AST analyzer, Git manager, Sentry tracing setup, and Ollama client.
• frontend/: React 18, TypeScript, Tailwind CSS, Vite, and Lucide icons.
• fixtures/: Test repositories for validating dead-code pruning and AST syntax safety.

---

How I Built It

OutOfOffice AI is built from the ground up around open-source AI and observability:

1. Local Intelligence with Google Gemma 2

The core reasoning engine runs Google Gemma 2:2b locally through Ollama (http://localhost:11434). The agent parses the target repository's Abstract Syntax Tree (AST), identifies unused imports or unreferenced functions, and formulates multi-step execution plans completely offline.

2. Distributed Telemetry with Sentry Agent Tracing

Autonomous agents shouldn't be black boxes. We instrumented the entire Python agent using Sentry's distributed tracing SDK. Every node in the execution graph is tracked with microsecond span precision:

• PlannerNode: Latency and prompt token metrics.
• ASTScanNode: File manifest and syntax structure inspection.
• GitBranchIsolation: Creation of isolated branches (ooo/agent-audit).
• VerificationNode: AST validation preventing syntax errors before patch emission. ### 3. Hands-Free Voice Debrief with ElevenLabs When returning to your desk, you don't need to dig through thousands of lines of terminal output. The backend synthesizes an executive voice summary using the ElevenLabs Turbo API (eleven_turbo_v2_5), providing a hands-free report of codebase health, modified files, and test results.

---

Why Does Open Innovation Matter?

Open innovation is essential to the core value of OutOfOffice AI:

• True Data Privacy: Closed cloud APIs require uploading private source code and internal architectures to remote servers. With open-weight models like Google Gemma 2, your proprietary code never leaves your local machine.
• Zero Token Cost at Scale: Autonomous coding loops perform multiple iterations—scanning dozens of files, verifying ASTs, and testing patches. On metered cloud APIs, this becomes cost-prohibitive. Open-weight models running on local hardware cost $0.00.
• Offline Resilience: Developers can work on planes, trains, or trails with zero internet connection. Open-source AI transforms intelligence from a metered subscription into durable, local computing infrastructure.

---

Prize Categories

I am entering the following categories:

• Best Use of Gemma ($200): OutOfOffice AI uses Google's open-weight Gemma 2:2b running locally via Ollama as the core intelligence engine for AST analysis and code refactoring.
• Best Use of Sentry Agent Tracing ($100): Full end-to-end distributed tracing across all agent reasoning nodes and tools, with microsecond span waterfalls and live telemetry.
• Best Use of ElevenLabs ($100): Synthesizes an executive hands-free audio debrief of codebase health and completed tasks, eliminating the need to read terminal logs when returning to the desk.