// the take
ADV.JS is a Markdown-based text-adventure/visual-novel engine aimed at AI-agent-driven authoring: you describe a game, an agent (Codex/Claude Code/Cursor) scaffolds it via CLI, you polish it in a local editor, then build and deploy to Cloudflare Pages. It's for developers building narrative games or wanting to embed a lightweight dialogue runtime into an existing web game, not for non-technical writers.
The compiler/runtime/client split is genuinely modular — you can compile .adv.md at build time and load just the runtime plus a scene JSON at runtime inside a host game, reusing your own UI, input and save system instead of the whole engine. The CLI is built for agent workflows end to end (adv init, adv agent install --mcp, adv check --json), with structured JSON output meant for tooling rather than humans. It also has real cross-platform ambition backed by tests: a Swift port of the engine on iOS with golden/parity tests comparing output against the JS engine, not just a wrapped WebView.
This is a huge monorepo for a 328-star WIP project — core engine, Electron desktop app, native iOS app with its own Swift engine, a CloudBase backend with audio services and billing-adjacent 'quota accounts', and a mobile SaaS (Studio) all live in one repo, which is a lot of surface for whatever team is maintaining it. Embedding is explicitly unstable: the docs say the lightweight runtime sub-entry 'has been implemented in the current workspace' but warns you to check the target package's exports and pin a verified version yourself before relying on it. Browser support is Chromium-only at launch (Firefox and Safari aren't in the test matrix), and the whole pipeline leans on pinned exact versions of the CLI, MCP server and Wrangler, which suggests the APIs are still moving underneath you.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →