AI / ML
Face-Recognition
Face detection, encoding, and recognition · OSS-backed by ageitgey/face_recognition (dlib ResNet, 99.38% LFW accuracy)
Ship production-ready stacks in minutes. Curated infrastructure as code, pre-configured for scale and refined with Japanese minimalist precision.
Face detection, encoding, and recognition · OSS-backed by ageitgey/face_recognition (dlib ResNet, 99.38% LFW accuracy)
Generative text-to-image · FLUX.1-schnell turns a text prompt into a high-quality image in 1 to 4 diffusion steps. Apache-2.0 licensed, fast, runs on a single L40S-class GPU.
Generative text-to-image · Stable Diffusion XL produces detailed 1024px images. Open weights, broad commercial use.
Generative text-to-image · Stable Diffusion 3.5 Large · strong prompt adherence and typography. Community License (free under $1M revenue).
Generative text-to-video · Wan 2.1 (1.3B) generates short clips up to 480p. Apache-2.0 · fully commercial, no restriction. Fits a single L40S.
Code generation · Qwen2.5-Coder 7B (AWQ) writes, explains, and refactors code across many languages. Apache-2.0 · fully commercial. OpenAI-compatible API · runs alongside other models on one GPU (capped VRAM).
Conversational chat · Qwen2.5-Instruct 7B (AWQ) for natural dialogue, small talk, Q&A, and general assistance across 29+ languages. Apache-2.0 · fully commercial. OpenAI-compatible · streams token-by-token · coexists with other models on one GPU (capped VRAM).
Image to 3D · TripoSR (Tripo AI + Stability AI) reconstructs a textured 3D mesh from a single image in seconds. Outputs GLB · view inline or download for Blender/Unity/Unreal. MIT license · fully commercial · EU-clean.
Image to 3D · Stable Fast 3D (Stability AI) makes game-ready meshes with clean UV-unwrapped topology and PBR material params in under a second. Outputs GLB. Weights HF-gated · needs an HF token. Community license.
Image to 3D · Tencent Hunyuan3D 2.1 produces the highest-fidelity textures of the open 3D models via a two-stage shape+paint pipeline. Shape-only ~6GB · shape+texture ~12-16GB. Outputs GLB. Tencent Hunyuan Community License (review for commercial use).
Reasoning · Gemma 4 26B-A4B (Google DeepMind) with built-in step-by-step thinking mode. MoE runs near 4B speed with 26B-class quality across math, code, and analysis. 256K context · 140+ languages. Apache-2.0 · fully commercial · EU-clean. OpenAI-compatible · streams.
Presentation generation · turns a prompt into a downloadable PowerPoint (.pptx). Asks a reasoning LLM (default: Gemma 4) to draft a structured deck, then renders real slides with speaker notes in the Kanso style. CPU-only · pairs with the gemma4-reasoning model.
African languages · AfroLlama V1 (Jacaranda) is an instruction-tuned Llama 3 8B specialized for Swahili, Yoruba, Hausa, isiZulu, isiXhosa, and English. Purpose-built for African-language chat, translation, and NLP where general models underperform. OpenAI-compatible · streams. Llama 3 community license (commercial-friendly).
Design image generation · Qwen-Image (Alibaba) creates UI mockups, posters, and infographics with best-in-class LEGIBLE text rendering (where most diffusion models fail). Wrap exact text in quotes in your prompt. Apache-2.0 · fully commercial · EU-clean.
Prompt optimizer (fast) · SuperPrompt expands a short or rough prompt into a rich, detailed one for text-to-image models (SDXL, Qwen-Image, Flux). A tiny 77M model that rivals 1B+ LLMs for prompt upsampling. Type 'a cat' and get a full descriptive prompt to feed your image generator. CPU-only · sub-second · Apache-2.0.
Prompt optimizer (high quality) · PromptEnhancer-7B (Tencent, CVPR 2026) rewrites your prompt into a clearer, structured, detail-rich image prompt using chain-of-thought, with specialized understanding of text-to-image failure modes (attribute binding, negation, composition). Higher quality than the fast SuperPrompt, at 7B / ~13GB VRAM. Pairs with SDXL, Qwen-Image, Flux, LTX, Wan.
Image to 3D Gaussian Splatting · TripoSplat (VAST-AI) turns a single photo into a native 3D Gaussian splat (.ply/.splat) in seconds, feed-forward with no optimization loop. Same team as TripoSR, but outputs Gaussian splats for Unreal/Unity/Blender and web viewers (SuperSplat) instead of a mesh. Adjustable Gaussian count. MIT licensed, fully commercial. Note: single-image means the unseen side is inferred, not reconstructed.
Zero-shot image recognition and classification · label any image against arbitrary text classes, no per-class training · OSS-backed by OpenCLIP
Zero-shot product recognition · identify or classify retail products in an image against your own catalog labels, no per-product training · OSS-backed by OpenCLIP (ViT-B/32). MIT licensed · fully commercial · CPU-friendly.
Production retail SKU recognition · detects every product in a scene (Faster R-CNN), fingerprints each with SigLIP 2, and matches against your enrolled SKU catalog · enroll via /enroll, recognize via /recognize · OSS-backed (torchvision BSD + SigLIP 2 Apache-2.0) · fully commercial
Self-hosted secure video conferencing. WebRTC-based with HD audio/video and end-to-end encryption.
High-performance WebRTC SFU for real-time audio, video, and data streaming at scale.
Low-latency voice chat for gaming and teams. Crystal-clear audio with sub-50ms latency.
GPU-accelerated robotics simulation. WebRTC-streamed scene viewer with full physics, sensors, and ROS bridge.
Self-hosted Semaphore UI for managing Ansible playbooks, inventories, and scheduled runs.
High-throughput distributed task orchestration. Scales compute workloads across a cluster of workers.
A full Firefox browser served over the web (KasmVNC). Browse from any device through a URL, with the session running on your machine.
A full Chromium browser served over the web (KasmVNC). Browse from any device through a URL, with the session running on your machine.
Open-source headless CMS with a Postgres database. REST + GraphQL API, admin panel, and content modelling. Ready for content migration and JSON export.
Battle-tested relational database with automatic failover, point-in-time recovery, and daily encrypted snapshots.
Sub-millisecond in-memory key-value store. Global replication and built-in persistence for high-traffic apps.
Document-oriented NoSQL database with horizontal sharding, replica sets, and JSON-native querying.
Drop-in MySQL replacement. Open-source with enterprise features like Galera clustering and ColumnStore.
World's most popular open-source relational database. Production-tested at scale with InnoDB storage engine.
World's most popular CMS. Drop a theme, build a blog or business site. Bundled with a MySQL backend and persistent storage.
Personal AI agent, self-hosted. Messaging, tools, skills and autonomous tasks.
Personal AI agent focused on self-improvement · persistent memory and a growing skill set.
General autonomous agent. Computer and terminal use, multi-agent delegation.
Lightweight personal agent written in Rust · low resource usage, quick to run anywhere.
Modern publishing platform for content creators. Clean editor, built-in newsletter, SEO-friendly themes. SQLite backend for easy setup.
Self-hosted file sync, calendar, contacts, and collaboration suite. Open-source alternative to Google Drive and Dropbox.
Workflow automation. Connect APIs, schedule tasks, build AI agents with a visual editor. Self-hosted alternative to Zapier and Make.
Self-hosted password manager · Bitwarden-compatible server. End-to-end encrypted vault that syncs across all your devices.
Lightweight self-hosted Git server. Issues, pull requests, mirroring, CI hooks · the GitHub experience on your own infra.
Create a free account in 30 seconds and we'll do the rest. Your selection is saved.