Skip to content

Security: ApeCodeAI/finsight

Security

SECURITY.md

Security policy

FinSight is a local-first, single-user tool. Your data lives on your machine (~/.finsight/) and your vault directory. There is no server-side component, no telemetry, no account system.

That said, a few security-relevant things still apply.

Reporting a vulnerability

Please do not open a public GitHub issue for security problems.

Email the maintainer listed in package.json with:

  • A clear description of the issue
  • Steps to reproduce
  • The FinSight version (finsight --version) and OS
  • Your assessment of impact, if any

You'll get a reply within a few days. If a fix is needed, we'll coordinate disclosure timing with you before publishing.

Scope

In scope:

  • Bugs that allow remote code execution via the local web server
  • Bugs that let a network neighbor read your portfolio when the dashboard is bound to a non-loopback interface
  • Connector code that mishandles untrusted upstream responses (e.g., quote feed returning malicious payloads)
  • Vault/ledger code that could be tricked into writing outside the configured ledger_dir

Out of scope:

  • "The web dashboard has no auth" — this is by design. The dashboard is meant to bind to localhost only. If you expose it to a network, put a reverse proxy with auth in front of it (Caddy + basic auth, Tailscale serve, Cloudflare Access). See the Security section in the README.
  • Issues that require you to run code from an untrusted source (e.g., finsight ledger restore against a malicious vault you cloned). FinSight trusts your local files.

Hardening tips for users

  • Keep your ledger_dir in a private git repo or an encrypted filesystem. It contains your full position list.
  • Don't pipe finsight context into a third-party LLM if you consider your portfolio sensitive — finsight context --json and the Markdown form both include account names and dollar amounts.
  • Don't expose finsight web on 0.0.0.0 without auth in front.

Thanks for helping keep FinSight users safe.

There aren't any published security advisories