[ ## ] // security

Security

ArcFlare is a local tool: the models, your prompts and your files stay on your machine. It is also an agent that can run commands, so here is plainly what it can do and what stops it.

this website

HTTPS only, strict headers

Served over HTTPS with HSTS, a content security policy, no framing, and no third-party scripts, trackers or ads.

No accounts, no analytics

There is nothing to sign in to and nothing that profiles you. See Privacy.

Remote control keys never reach the server

The /rc key travels in the link's #fragment, which browsers don't send. The relay only ever sees a SHA-256 hash of it, and sessions expire after six idle hours.

the cli and desktop app

It runs with your permissions, by design

The agent can run shell commands, edit files and (if you turn on the machine server) use the desktop. It is not a sandbox. Every command and edit asks first unless you switch on auto mode. For untrusted code, run it in a container or VM.

Project MCP configs are gated

A repository's .mcp.json never starts anything until you approve it. The prompt shows each command, each remote host, and every environment variable it would send. Any change to the file asks again.

Updates come from one place

arcflare update only installs from the official repository, shows the incoming commits, and waits for you to confirm. The background check only checks. It never installs anything.

Tokens stay local

OAuth tokens for hosted MCP servers live in ~/.arcflare/oauth.json, readable only by your user. arcflare mcp logout deletes them.

The desktop app is sandboxed

Its interface runs with no Node access and can only call a fixed list of actions. It can open only files the app made or that you picked.

found a vulnerability?

Please report it privately through a GitHub security advisory rather than a public issue. You'll get a reply within a few days. The full threat model is in SECURITY.md. Contact details are also in security.txt.