GMod Sandbox questions, answered

Sandbox is the default gamemode. It is built in, and the maps gm_construct and gm_flatgrass ship with the game. A gamemode of your own can build on it with DeriveGamemode("sandbox"), and Codelab Station can write that file for you to review.

It is writing Lua files that the game loads and runs, on the server, on a player's client or on both. Addons go in garrysmod/addons, each with a lua/ folder. Files in lua/autorun/ run on both sides. Errors appear in the console.

A realm is the place where Lua code runs: the SERVER or a player's CLIENT. Files in lua/autorun/server run only on the server, files in lua/autorun/client only on a client, and AddCSLuaFile() marks a file to be sent to clients. Mixing them up is common.

The usual cause is a missing AddCSLuaFile() call or a realm mistake. In singleplayer the game is both server and client on one PC, so a mistake that only shows on a dedicated server can stay hidden. Codelab Station reads the files with grep and read_file and shows a diff for the fix. Then retest on a server.

Yes, codelabss.com is the official Codelab Station website. Get the original build from the download page and compare its SHA-256 checksum with the published manifest before you run setup. Support is in the app. Questions go through the Send feedback button there.

No. A model can write Lua that fails, even when the diff looks fine. Shell reports the real exit code of your test command, and the game console prints Lua errors when you load the map. Treat every change as a draft and read it before you accept.

A gun addon is a scripted weapon, so start with the console, where Lua errors are printed with a file and line. The agent then checks lua/weapons/ with list_dir and grep, reads the file named in the error and proposes an edit_file diff you review. Then test again.

No. Codelab Station is not a mod manager in GMod, and it does not subscribe to, download or update Steam Workshop items for you. It edits the files in your working folder and runs your own commands, and you stay in charge of what is published.

A cloud AI provider through a standard API endpoint, a model running on your own PC or the built-in model. You choose. The Test button checks the endpoint and shows the response time and the context window, and prompts go only to the endpoint you picked.

Four things can use the network: prompts to the endpoint you pick, web search, control of chats from a phone browser and update checks. With a local model nothing leaves your machine. There is no telemetry. Chats and API keys stay on your PC.

The installer is not code-signed yet, so Windows SmartScreen may show a warning. Compare its SHA-256 checksum with the published manifest, and only after it matches choose More info and then Run anyway. If they differ, stop.