Verified project record

nexu-io/html-anything

A local-first, agentic HTML editor that uses auto-detected coding-agent CLIs and composable skill templates to turn raw text or data into designed, single-file HTML. It relies on existing CLI sessions and provides one-click exports to social platforms.

Project overview

It avoids requiring a separate API key by reusing locally installed, logged-in coding-agent CLIs to automate HTML layout and design tasks via streaming previews.

Project type: AI Agent · Prompt Engineering
Use cases: Documents & Office · Coding & Development
Deployment: Refer to project documentation
License: Apache-2.0

Best for

Creators and developers who want to convert raw text or tabular data into designed HTML layouts using locally installed, logged-in coding-agent CLIs.

Key capabilities

Scans the user's PATH on startup to auto-detect installed coding-agent CLIs—such as Claude Code, Codex, and Cursor Agent—and reuses their existing logged-in sessions.
Provides 75 composable skill templates across 9 deliverable surfaces, including examples such as magazine articles, keynote decks, and posters, to define the output format and design system.
Streams the agent's stdout JSON-line output via Server-Sent Events to live-update an iframe preview, allowing users to monitor the HTML render line by line and interrupt mid-generation.
Runs user-emitted HTML in an isolated sandboxed iframe with allow-scripts and allow-same-origin, quarantining cookies and localStorage from the host while supporting Tailwind CDN and Google Fonts.
Accepts Markdown, CSV, TSV, JSON, SQL, and plain text inputs, parsing tabular data directly in the browser without uploading anything.
Provides one-click export utilizing CSS inlining and high-DPI screenshots to WeChat, X, Zhihu, and standalone HTML or PNG files.
Planned feature to generate four candidates from the same brief and allow the user to select one.

Limitations and risks

Does not support direct video rendering; Hyperframes frame scripts must be handed off to Remotion separately to render an MP4.
The local API surface uses maximally permissive flags to spawn coding-agent CLIs, which requires Host-header allowlists to prevent DNS-rebinding attacks.

Getting started

Documented as an easy setup requiring 4 commands: clone the repository, navigate into the html-anything directory, run pnpm install, and execute pnpm -F @html-anything/next dev before opening http://localhost:3000.

Alternatives and comparisons

A local-first, open-source desktop app that turns prompts into prototypes and assets using your existing model provider.
Ready-to-use markdown skills that provide AI agents with frameworks for marketing tasks like CRO, copywriting, and SEO.
Rewrites text to remove signs of AI-generated writing based on documented content, language, and style patterns.

Project comparisons

Evidence and sources

GitHub project description: The agentic HTML editor — your local AI agent writes the HTML, you ship it.
README: Local-first, zero API key, reuses the CLI session you already have logged in — **9 coding-agent CLIs** auto-detected on your `PATH`
README: "writing HTML" used to mean writing CSS, picking type scales, snapping to a grid, doing responsive — most users won't, designers won't bother, writers don't have the patience.
README: - [ ] **Multi-template compare preview** — generate four candidates of the same brief, pick the best one
README: | **Claude Code** | `claude` | `claude -p --output-format stream-json` | | **OpenAI Codex** | `codex` | `codex exec --json --sandbox workspace-write` |

Reviewed at: 2026-07-18T05:55:27Z

Open original GitHub project