archify-hai

One-line pitch. Turn a codebase or system description into a polished, interactive system map โ€” directly in chat.

Archify is a deterministic Node.js rendering and validation system for coding agents (Cursor, Claude Code, Codex CLI, OpenCode). Agents emit typed JSON IR; Archify compiles it into self-contained HTML/SVG with five diagram types, four presets, dark/light themes, and finite motion.

This Hugging Face mirror is maintained by smlflg โ€” same project as tt-a1i/archify on GitHub.

Why this mirror?

Archify is part of a broader research line called HAI โ€” Human-Agent Interface: the engineering of how autonomous coding agents present their work to humans in a verifiable, navigable form. The HF Hub is the natural home for:

  • Versioned model cards that document Archify's IR contract
  • Stable URLs for proof artifacts (e.g. MCO runtime architecture, release-delivery workflow) that survive site redesigns
  • Discoverability for agent developers evaluating tooling

Quick start

npx skills add tt-a1i/archify -g

or grab the offline bundle and drop it into your agent's skills directory:

hf download smlflg/archify-hai --include "archify.zip"

The complete generation and viewer contract lives in archify/SKILL.md.

Five diagram types

Type Use case
Architecture System topology, components and edges
Workflow Authored happy paths with chapters
Sequence Cache-miss / event-stream time-ordered traces
Lifecycle Long-running agent runs with states
Dataflow Data movement between roles

HAI mode (Human-Agent Interface)

Archify treats every map as an auditable claim set โ€” not a picture. Each artifact exposes:

  • Stable links โ€” #focus=<id>, #route=<src>~<dst>, #lens=<role>~<role>, #view=<view-id>
  • Revision verification โ€” JSON IR is hashed; viewer can prove the page reflects the checked-in source
  • Finite motion โ€” reader-driven, respects prefers-reduced-motion, never enters canonical exports
  • Delta review โ€” compare two validated snapshots as Before / Delta / After with exact added, removed, changed, moved, and rerouted facts

This is the HAI promise: an agent's output must remain inspectable by a human without losing intent or context.

What's in this repo

Path Purpose
archify/ Core renderer, schemas, examples, and viewer (the Skill payload)
docs/ Documentation site (HTML pages, gallery, asset images)
examples/ Generated HTML artifacts + JSON IR sources
benchmarks/ Validation suite for the renderer contract
experiments/ R&D branches (Mermaid validation, visual evolution, showcases)
integrations/ DeepSeek-harness plugin wrapper
scripts/ Dev helpers (build, lint, validate)
generated/ Snapshot of the live Maka demo workflow
README*.md, DESIGN.md, ROADMAP.md, CHANGELOG.md Project meta-docs (EN / DE / ZH)

Owner

Maintained by smlflg โ€” CS student, KIยฒNG program (TH Mannheim); part-time Werkstudent in Berlin; builder of the HAI stack and several agentic harnesses (HAI-Agent-Workflow, HAI-DPPM, Hermes-Agent).

License

MIT โ€” see LICENSE.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support