Useful AI tools, practical use cases and open models.
See why they matter and find something to try.
The Moss Keeper is an open-source project showcasing how procedural code generated by frontier models can produce an entire Studio Ghibli-style anime short within a single dependency-free HTML file.
Instead of using heavy video generation models or external cloud CDNs, this demonstration embeds custom canvas film-grain shaders, audio controls, and procedural rendering entirely in self-contained offline HTML.
Jev Agent introduces a separate, fast tool-choice step that picks the winning tool from up to 255 options before passing context to an LLM to generate tool arguments.
In agents with dozens of available tools, feeding full schemas into an LLM on every step increases latency and token costs. Decoupling tool selection from argument generation cuts round-trip time on multi-step workflows.
AgentWeb provides an HTTP and MCP endpoint for AI agents to query structured local-place data, such as cafes, shops, and business records, without running a browser scraper.
When building local research or travel workflows, running headless browsers or scraping unstructured search results is fragile and slow. A clean JSON tool call gives the agent consistent fields like name, address, and website directly.
Images Upscale is a browser-based utility running Real-ESRGAN models to upscale JPG, PNG, and WebP images by 2x or 4x without watermarks or mandatory sign-ups.
Builders and creators frequently need to enlarge small UI screenshots or product images for documentation, landing pages, and pitch decks. It offers 2 free daily credits without requiring a paid subscription or software installation.
Selvedge is an open-source MIT-licensed Python CLI and local MCP server designed to persist architectural decisions, rejected approaches, and code context across AI coding sessions.
When working with coding agents like Claude Code, Cursor, or Codex, context often gets lost when chat sessions reset or servers restart. Selvedge provides local, searchable decision memory without sending code to a third-party cloud.