Clients, gateways, workers, and databases pile into undifferentiated boxes that collapse past 15 nodes.
Let AI read your real repo and draw true spatial architecture.
Equip your AI coding agent to read real repo source and generate interactive 3D & 2D spatial architecture maps in seconds.
- Claude Code
- Cursor
- Antigravity
- Codex
- Pi
ArchRay in action: live single-file HTML.
Top strip shows the coding-agent prompt; below embeds the real self-contained FreshMart HTML build.
Inspect any layer—view all 10 layers or combine ends freely.
Inspect all 10 layers with one command, focus on frontend/backend/data, or combine multiple layers with commas; 1-hop neighbors stay softly visible.
> /archray all
Prompt: “Use archray to inspect source code and build the full 3D map” · CLI: --layers all --style tech|blueprint|glass|paper
Full 10-Layer System Overview
Inspect repository source once to map clients, gateways, central services, async/AI wings, external SaaS orbits, and storage/infra plinths.
Explore this layer in the live sandbox ↓10 layers & combinations Open to select one layer or combine several
Detects and removes architecture diagram slop.
One switch turns all five specimens from generic AI failure modes into deterministic ArchRay answers.
Without schema validation or file:line citations, pretty diagrams silently drift from real code.
Zero visual distinction between stateful persistence, async workers, and third-party SaaS APIs.
Once exported, static charts offer zero help for tech-debt triage or migration planning.
Viewport captures chop off outer layers while SaaS diagramming uploads proprietary repo metadata.
Built around your real engineering repo, not a toy demo.
Everything here comes from reading real source repositories: toy diagrams draw 4 unlinked boxes without citations, while ArchRay is benchmarked on 7,700+ file monorepos with measured vs. ≈ estimate provenance.
Dense packages auto-collapse (+N) into a 196ms ELK layout with 60fps 3D orbit.
apps/web · packages/features · 196ms
Finds the 6.2k LOC sync lock hotspot and writes Passport → Clerk migration prompts.
apps/api/src/order/order.service.ts
Maps App Router, Prisma, and Stripe flows while flagging hardcoded JWT_SECRET fallbacks.
middleware.ts · prisma/schema.prisma
Agents extract facts. The engine computes geometry.
LLMs never hand-author SVG coordinates. Your agent outputs verified modules and source evidence; deterministic layout and 3D projection stay clean inside .archray/.
Every claim lives in auditable repo JSON.
Architecture state stays in Git-friendly .archray/graph.json and view.json, separating measured facts from ≈ estimates with unclipped PNG & QR poster exports.
Install ArchRay
Install into your project root in 30 seconds, open a fresh agent session, and run /archray.
- 01Does not modify application source
- 02CLI, viewer and Skill are self-contained in .archray/tool
- 03Keeps graph/view/index; update or uninstall anytime
Step 1: Copy install command & run in repo root
$ curl -fsSL https://github.com/loofare/ArchRay/releases/latest/download/install.sh | sh -s -- . claude
Step 2: Once installed, copy & say this to your AI
Map this project architecture with archray (or type /archray)
Health check, update & clean uninstall:
$ ./.archray/archray doctor && ./.archray/archray update --check
FAQ
Short answers about ArchRay capabilities, data boundaries, hosts and updates.
What is ArchRay?
ArchRay is an AI-first architecture visualiser. Install it into a repository, type /archray in your AI assistant, and it reads real source to produce layered, interactive, offline 3D/2D single-file HTML.
Does it upload my source?
No. Analysis runs inside the AI assistant session you authorize, and artifacts stay in the local .archray directory; the HTML is self-contained. Only an explicit update contacts GitHub Releases to read the version and verify the archive.
Which AI hosts are supported?
Claude Code, Cursor, Antigravity, Pi and Codex. Claude/Cursor/Pi use /archray; Codex uses $archray or /skills. Install with --agent all or a comma-separated list.
How do I install and update?
Download the release or run the bootstrap install.sh in the target repository root. Then use ./.archray/archray update --check followed by update; v0.4 and older need a one-time bootstrap upgrade.
How is it different from a generic AI diagram?
ArchRay separates semantic facts from deterministic layout: the AI records source evidence and dependencies, while the layout engine computes coordinates and 3D/2D projections. It focuses on traceable paths, offline delivery, layers, health and replaceability instead of a one-off image.
Which repositories is it for?
It fits repositories where you need to understand real layers, cross-layer dependencies, technical debt or migration options, especially when the map should be shared offline. It is not runtime monitoring or an automatic scanner; description-only maps need extra review.
Privacy Policy & Collaboration
Your source code never leaves your machine. Reach out directly for partnerships, custom workflows, or feedback.
ArchRay Privacy & Data Security Policy
- 1. 100% Local Execution · Zero Source Upload ArchRay CLI and single-file HTML views run entirely offline on your machine. No source code, file paths, secrets, or architecture graphs are ever uploaded to archray.dev or external servers.
- 2. Host Coding-Agent Boundary Source analysis happens solely inside your own authorized AI coding session (Claude Code, Cursor, Antigravity, Pi, or Codex) under your chosen provider’s terms.
- 3. Zero Telemetry, Zero Cookies, MIT Auditable Generated HTML views and CLI tools contain zero analytics beacons or tracking cookies; update checks perform read-only GitHub Release SHA-256 verification.