Questions
Short answers about what ModScrim does, what it never does, and what happens to your data.
For streamers
Does ModScrim post anything to my chat?
No. Attacks only exist in the practice chat. The one optional exception is a single fixed line when a drill starts: “ModScrim practice drill starting. Practice chat only, nothing here is real.” It’s off unless you turn it on.
Is the grading real?
On a real channel, yes. Twitch’s Check AutoMod Status grades every line it can within your channel’s hourly limit. Anything graded by ModScrim’s own matcher is labelled “local estimate”, and demo rooms always use the local estimate.
Can ModScrim change my AutoMod settings?
No. It reads them. Twitch’s update call replaces every setting at once, so ModScrim never uses it. The only thing it can change is your blocked terms, and only after you press Approve and push.
What if a new term bans a regular?
ModScrim rejects any term that matches your last 7 days of chat, and push is disabled unless real-chat matches are zero. If one still causes trouble later, Undo this push removes it, or you can remove it in your Twitch dashboard.
How long does a drill take?
The drill itself runs 30 seconds to 3 minutes. From the first whistle to the report is usually under six minutes.
Do I need to install anything?
No. You, your mods and OBS all use links.
How much does it cost?
ModScrim is a free, open-source hackathon project. It runs on Cloudflare’s free tier and makes no paid AI calls.
For mods
Do my actions affect real viewers?
No. Delete, time out and ban in the practice chat only score the drill.
Can I use a phone?
Yes. The mod view works from 360 pixels wide, and on a phone every button is at least 44 pixels tall. On a keyboard, D, T and B act on the focused line.
Will my name show on stream?
Your display name appears on the overlay scoreboard unless your streamer turns on Hide mod names, which shows Mod 1, Mod 2 and so on.
Data and safety
What happens to my chat data?
It stays in your channel’s own storage for 7 days, is used only for the false-ban check, and is deleted when you ask, sign out or revoke the app. The privacy policy has the details.
Why not use an AI model to write attacks?
Templates and tricks are safer, free and repeatable: the same seed always gives the same wave, so a drill can be replayed exactly and tested.
What can’t blocked terms catch?
Spaced or dotted letters, zero-width characters, emoji inside a word and look-alike letters. ModScrim labels those “needs a bot rule”.
Can viewers learn tricks from the overlay?
No. The overlay never receives attack text, only the length of each line, drawn as a blurred bar.
About the project
Is ModScrim made by Twitch?
No. It’s an independent project that signs in with Twitch and uses Twitch’s public API. It isn’t affiliated with or endorsed by Twitch.
Where is the code?
In the public repository. It was built with OpenAI Codex for the Codex x TwitchCon hackathon.
