Please don't open a public issue for a security problem. Report it privately through GitHub's private vulnerability reporting on this repo.

Say what you found, how to reproduce it, and what it lets someone do. We'll acknowledge the report within a few days and keep you posted while it's fixed. We're glad to credit you in the release notes if you'd like.

• Lamp.app itself, including the local HTTP endpoints it runs on your Mac (approvals, the agent's tools, the AI gateway) and the way it signs and installs the apps it builds.
• Generated apps: the files Lamp writes into every app (LampSupport.swift and friends, see Sources/Lamp/Template.swift), and how an app talks to Lamp and to other apps on your local network.
• The website and update service in site/ and updates/ (lamp.dev and updates.lamp.dev): bug reports, shared wishes, the update feed and downloads.

Bugs in the coding agents Lamp bundles or drives (pi, Claude Code, Codex, llama.cpp) belong with those projects, unless the problem is in how Lamp runs them.

Lamp runs on your Mac, and so does the agent. Lamp's own commands over an app's folder run in that app's sandbox. Whether the agent's shell commands wait for your approval is a setting (Settings → Building: Only when risky, Always or Never); with Only when risky, lower-risk commands run in the sandbox without asking. Files outside an app's own folder are checked the same way whichever engine is building.

API keys stay on your Mac, in the login Keychain (release builds) or a private file (builds from source), and are sent only to the provider they're for, with one exception: an iPhone or iPad app that uses AI away from your Mac is given your OpenRouter key, which it keeps in the phone's Keychain. Help → AI & Privacy inside Lamp says all of this to the person using it.