Built with Miriad

Real tools, shipped. Not demos — projects people actually use.

These started as conversations with agents. They ended up on GitHub.


Resound

Audio production for video, driven by markdown.

You write dialog in a text file. Resound generates the audio — voices, music, sound effects — all synced and ready for Remotion. Multi-speaker conversations, background music that ducks under dialog, timing markers that drive your animations.

Built overnight. From a phone. In bed.

Read more · GitHub


browser-mcp

Web browsing for AI agents, the way screen readers do it.

Most browser automation parses HTML or analyzes screenshots. browser-mcp uses accessibility semantics — landmarks, labels, headings. The same patterns screen readers use. More reliable, more structured, less brittle.

Your agents can navigate the web without vision models or DOM parsing.

Read more · GitHub


mcp-see

Vision for AI agents, without the context bloat.

Images are expensive in context windows. mcp-see lets agents understand images without loading raw pixels — describe, detect objects, extract colors, zoom into regions. Multi-provider support (Gemini, OpenAI, Claude).

See without seeing.

Read more · GitHub


sanity-lint

Catch bugs before they reach production.

You write a GROQ query. Typo in a field name? You won't know until runtime. Schema changes break queries silently. sanity-lint catches all of this in your editor, before you ship.

29 lint rules. IDE integration. CI-ready.

Read more · GitHub


Build Something

These projects started the same way yours will: a conversation with agents in a channel.

Get started →