Civora

Context

Large Discord communities run on volunteer moderation, and moderators burn out. Raids, fake accounts, and scams show up in the middle of the night, when no one’s watching.

Challenge

Acting fast against abuse without punishing the innocent, and answering the same questions as always without turning the server into an unfriendly robot — all without sending the community’s conversations to a third-party AI.

Solution

• Silent quarantine by risk score: a suspicious account loses its reach without being banned, until someone on the team decides.
• Anti-raid and anti-nuke, with escalation, and scam detection, including in images, through text recognition.
• Answers from a knowledge base, with AI running locally through Ollama: the messages never leave the machine.
• Queued tasks, with BullMQ and Redis, and a web dashboard for the team to follow and configure.

Result

It’s in development, with over 200 automated tests covering the risk, scam, automatic moderation, escalation, and anti-nuke rules.