A self-hosted agentic coding assistant that runs a local 7B language model in a tool-use loop — reading, writing and editing files, running shell commands, and streaming progress in real time. No external APIs, no usage limits, runs entirely on private hardware.
GarmentScan is a Telegram bot that photographs a garment or care label and calculates the full environmental lifecycle: CO2 emissions, water usage, bill of materials, production cost, EU Digital Product Passport preview and a Greta Mode sustainability rating.
AutoValuta is a Telegram bot for Italian used car buyers. Photograph a license plate and it identifies the vehicle via the InfoTarga registry, scrapes AutoScout24 and Subito.it for comparable listings, checks insurance status, and gives you a GOOD VALUE, FAIR or OVERPRICED verdict in seconds.
DocTranslate uses a vision-language model to deconstruct any document image or PDF into its structural components, then translates every element with a 72B LLM — preserving headings, tables, lists and two-column layouts in clean HTML output.
A self-hosted AI music generation studio built on ACE-Step, with LLM-powered prompt engineering, eight genre presets from deep house to drum and bass, and full control over BPM, duration and guidance scale.
CrateVision is a Telegram bot for vinyl crate diggers. Photograph a record at a flea market, and it identifies the release using computer vision, checks Discogs and Vinted prices, pulls BPM and sample connections, then gives you a BUY, MILD or SKIP verdict in seconds.
Cloud AI image tools come with content restrictions, ownership ambiguity and no control over the models being used. Here is why serious artists, illustrators and creative studios are moving to self-hosted generation — and what that looks like in practice.
A look inside the LoRA Playground — a self-hosted AI image generation tool built on Stable Diffusion, with LoRA model support, LLM-powered prompt enhancement, eight style rewrite presets and a persistent gallery.
PRAG (Private RAG) is a self-hosted AI knowledge base system that lets companies query their own documents, manuals, contracts and reports using natural language — with no data ever leaving their infrastructure.
A technical deep dive into a production RAG (Retrieval Augmented Generation) system built for Italian legal document analysis — covering hybrid search, multi-LLM support, and a GPT-4o Oracle mode that automatically detects and corrects hallucinated answers.