Cockatoo
An experimental research platform exploring neural networks and machine learning for real-time Discord moderation and threat analysis.
An experimental research platform exploring neural networks and machine learning for real-time Discord moderation and threat analysis.
A few ideas we're exploring
Instead of relying on blanket bans of URLs or invites, Cockatoo tries to understand the context and decides if the content is actually harmful in real time.
We only keep anonymized threat intelligence data on-premise, nothing more. We do not store or log messages or user information on our side, ever.
We can analyze the full suite of content from URLs to attachments, providing a unified threat intelligence platform.
See how our experimental approach stacks up against traditional keyword-based filtering systems.
Static keyword filters block harmless content
Rigid rules often block things they shouldn't
Context-aware filtering that actually understands
Analyzes context to allow safe content while catching actual threats
Real-time analysis tries to understand context, not just match keywords.
Cockatoo comes in two variants
The experimental version that uses neural networks and machine learning to adaptively detect threats. This one does real-time analysis and categorization using ML models.
The open-source version that uses the public dataset from Core to block known threats. You can self-host this on your own infrastructure. It's lighter since it doesn't do real-time ML analysis.
The Edge version and the dataset are available on GitHub if you want to check it out or contribute. All our research, models, and tools are open-source for reuse and adoption.
Questions or feedback? Just squawk at us!