This is my submission for the Hacktoberfest Open-Source AI Challenge — Week 3: Slow and Steady.
What I Built
SlothDesk AI is a local-first productivity assistant that helps developers analyze documents, extract action items, identify duplicate files, organize workspaces, and profile CSV datasets.
Instead of letting an AI agent make uncontrolled filesystem changes, SlothDesk AI creates reviewable proposals and requires explicit approval before executing supported operations.
Key features include:
- Document analysis for TXT, Markdown, CSV, and JSON files.
- AI-assisted summaries, tasks, decisions, and deadline extraction when local Gemma inference is available.
- SHA-256-based duplicate detection.
- Dry-run proposals, explicit approval, execution history, and rollback safeguards.
- CSV data profiling for missing values, duplicate rows, and inferred column types.
- Deterministic fallback behavior when local AI inference is unavailable.
The goal is simple: spend less time on repetitive file management and more time building things.
Demo
Live application: https://slothdesk-ai-o4mk.onrender.com/
The hosted demo runs in a restricted mode to protect users' local files. Local filesystem workflows and Gemma inference can be tested by running the project locally with the required dependencies and model installed.
Code
GitHub repository: https://github.com/dr5t/SlothDesk-Ai
The project is built with TypeScript, Node.js, and a responsive HTML/CSS interface. The repository includes the source code, tests, documentation, and deployment configuration.
How I Built It
SlothDesk AI combines conventional software engineering with open-weight AI inference.
Gemma 3 + Ollama: The application integrates Gemma 3 (gemma3:4b) through Ollama for supported document-analysis workflows. When the model is available, its structured output can supply summaries, tasks, decisions, deadlines, and document categories. When inference fails or the model is unavailable, the application falls back to deterministic analysis and identifies the provider used.
Human-in-the-loop automation: Analysis and execution are separate steps. The application presents proposed file operations for review and requires explicit approval before supported changes are made.
Filesystem safety: Workspace boundary checks, symlink protections, collision-safe destinations, SHA-256 duplicate detection, and hash-verified rollback help keep file operations controlled.
Testing: The automated test suite passes 68 out of 68 tests, and the TypeScript build completes successfully. These tests include mocked Ollama responses, timeout handling, invalid-response fallback, and filesystem safety checks.
Deployment: The application is deployed on Render. The hosted deployment uses restricted mode rather than claiming access to a user's local Ollama runtime or filesystem.
Why Does Open Innovation Matter?
AI productivity tools should not force every document through a proprietary cloud API.
Open-weight models and local inference give developers more control over where supported AI processing happens, which model they use, and how they adapt the system to their needs. With Ollama and Gemma, developers can experiment with document analysis on their own machines without making a closed hosted model the only option.
Open-source code also makes the safety mechanisms inspectable. Developers can examine how proposals are generated, how approval is enforced, and how rollback verifies file integrity.
For SlothDesk AI, open innovation means making AI-assisted productivity more transparent, adaptable, and user-controlled.
My Agent Session
[Add your DevRelay agent session link here, if available.]
The Lazy Payoff
SlothDesk AI reduces repetitive work by bringing document analysis, task extraction, duplicate detection, and CSV profiling into one reviewable workflow.
Rather than manually inspecting every file or making unreviewed changes, users can start with structured findings and decide what to do next.
I have not yet established a reliable measured time-saving figure, so I am not claiming a specific number of minutes saved. The aim is to make routine work easier to review and safer to automate.
Prize Categories
- Best Use of Gemma — for the Gemma 3 integration through Ollama in supported local document-analysis workflows.
- Best Use of Render — for deploying the application on Render.
These are the categories I intend to enter, subject to the official challenge rules and eligibility requirements.
Built for developers who want to be a little lazier about repetitive work and a lot more productive about building.
Top comments (0)