The Concept Model app is a local-only dev tool. It is designed to run on your own machine, alongside Claude Code or another local AI assistant. It is NOT designed for:

• Multi-user deployment
• LAN exposure
• Internet-facing deployment
• Storing sensitive / regulated data

The dev server binds to 127.0.0.1:5116 and the SvelteKit +server.ts API routes write JSON files directly to data/. There is no authentication. Anyone who can reach the port can read, overwrite, or delete your concept models.

If you need to share a concept model, use the Export menu (JSON, CSV, Excel, SVG, Turtle or RDF) and share the downloaded files, never the server.

Only the latest commit on main is supported. We don't backport security fixes to older releases. If you're running a fork, rebase onto current main to pick up patches.

Please report security issues privately to security@AgileDataGuides.com. Include:

• A description of the issue and the impact you believe it has.
• Steps to reproduce, or a proof-of-concept if available.
• The commit SHA you tested against.

Do NOT open a public GitHub issue for vulnerability reports — opening one accidentally is fine, just close it and email us instead.

The following are intentional design choices, not vulnerabilities:

• No authentication. The app runs locally, so there's no user to authenticate. It is not built for multi-user access.
• CSP 'unsafe-inline'. Required by SvelteKit hydration and Tailwind. Removing it breaks the app.
• Filesystem writes from the dev API routes. The threat model assumes you trust your own machine. The API rejects requests that declare an oversize body, structurally invalid models, and path-escape attempts, but it doesn't try to defend against an attacker who already has shell access.
• data/*.json file format. Anyone with shell access can hand-edit the JSON files. That's by design — it's also how Claude Code reads your concept models.

We do enforce some hardening even within the local-only threat model:

• The dev server binds to loopback (127.0.0.1), never 0.0.0.0. Do NOT change vite.config.ts server.host.
• The POST and PUT routes reject requests whose Content-Length header declares more than 5 MB, and reject non-JSON or structurally invalid models. A request with no Content-Length header skips the size check.
• File paths from URL params are filtered through a kebab-case regex + path-prefix check so ../../etc/passwd-style traversal can't escape data/.
• We don't ship the abandoned xlsx package. Writing XLSX is done via exceljs; reading XLSX is intentionally unsupported (there is no safe parser of comparable scope).
• Dependabot watches package.json and the GitHub Actions workflows weekly for vulnerable updates.