Heartmorrow/DSim is a local-first, single-player application. The web client talks to a local Fastify server, and that server talks to the LLM endpoint configured by the user. This policy explains how to report security issues and how to run the project safely.

Security fixes are made on the default branch for the current development version. This project is still pre-1.0, so older commits, local forks, and unmaintained branches are not guaranteed to receive backported fixes.

When tagged releases are introduced, the latest release and the default branch are the supported targets unless stated otherwise.

Please do not open a public issue with exploit details, proof-of-concept code, or private data.

Use GitHub private vulnerability reporting, or reach out on the Discord and we can take a look at your report to triage and fix it.

Include as much of the following as you can:

• Affected commit, branch, or release.
• Operating system, Node.js version, and pnpm version.
• Whether the API server was bound to 127.0.0.1, localhost, a LAN address, or a public interface.
• Clear reproduction steps.
• The impact you believe is possible.
• Any relevant logs, stack traces, sample files, or requests.

This is a small project, but we'll try our best to do the following:

• Acknowledge a valid private report within 7 calendar days.
• Confirm the affected versions or commits.
• Share whether the issue is accepted, needs more information, or is considered out of scope.
• Credit the reporter if they want credit and the disclosure is coordinated.

Reports are especially useful when they affect:

• Arbitrary code execution, command execution, or unintended file writes.
• Path traversal or access outside configured data/upload directories.
• Secret exposure, particularly the LLM API keys stored in .env or local SQLite data.
• Unsafe handling of uploaded images, imported save bundles, or shared character/world packs.
• Zip bombs, archive traversal, symlink, or resource-exhaustion attacks in import paths.
• Cross-site scripting or browser-side data exfiltration, especially if reachable through imported data.
• Documentation that would lead users to expose the local server API unsafely.

The following are usually not security vulnerabilities by themselves:

• Model jailbreaks, prompt injection, or unwanted roleplay/content that does not escape the app's local data and execution boundaries.
• Claims that require the attacker to already control the user's machine, shell, repo checkout, browser profile, or filesystem.
• Denial of service from intentionally enormous local saves or data files created by the same trusted user, unless it bypasses an intended import/upload limit.
• Issues caused by deliberately binding the API to a public interface.
• Vulnerabilities in third-party LLM servers, model providers, or browser extensions.

If in doubt, report privately anyway. We can deal with false alarms.

Heartmorrow is designed for one trusted local user, not for hosting as a shared internet service. There is currently no authentication, authorization, or multi-user account boundary. It is not designed to be used by anyone but a single user at this time.

If you decide to host this publicly somewhere, even if you intend for it to be private, please be aware that at this time this is not an intended configuration. This may change in the future.

When using a remote or cloud LLM endpoint, prompts, character data, conversation context, uploaded images sent to a vision model, and other generated context may leave the machine and be processed by that provider. Review the provider's data retention and privacy terms before using private content.

For local LLM servers, prefer loopback URLs such as http://localhost:1234/v1 or http://127.0.0.1:1234/v1. Avoid pointing the app at untrusted endpoints. While LLM serving endpoints should not be able to influence the game in a security-relevant way no matter what they return, if you find an instance where this is not true, please let us know!

Uploaded images and imported share packs are untrusted input. The application attempts to limit risk with MIME allow-lists, size caps, generated filenames, path checks, strict JSON validation, and bounded ZIP decoding. Please still use normal caution:

• Import packs only from people or communities you trust.
• Avoid running third-party scripts, experimental plugins, or modified installers from untrusted sources.
• Keep backups of the data directory (/apps/server/data) before importing large or unfamiliar content.