Neuro-adaptive profiles
Choose any combination. They layer together.
A send-side filter that reads the conversation
The Mirror scans messages before they are sent — it does not scan what arrives.
It receives data from the connector and compose layer, not from you.
No paste box, no input field.
DysPhilter is active — messages are being filtered as they arrive.
No compose area. The engine receives from the connector layer, not from the keyboard.
Three layers, read together
Language
The actual words used.
Tone
How those words are being used — banter vs hostility, care vs control.
Inference & subtext
Meaning beneath the surface — words that look fine literally but carry harmful intent.
Bidirectional — context decides
Bank agent writes: “We require remediation of the arrearage within 14 days.” You receive: “You've missed a payment. You have 14 days to sort it — here's how.”
You write: “argh the money thing I can't, it's too much, please help.” The agent receives: “I'm struggling with my payment and would like to arrange a manageable plan.”
Context, not the words alone: “You muppet 😂” between mates bantering = safe, let through. The same words in a hostile exchange = stopped.
If both send hostile content at once
You
Idle
The other person
Idle