A middleware layer
for AI.
Anchored. Tethered. Awake.
A working middleware that sits between an operator and a connected LLM and enforces a Three-Layer Alignment Stack: an Origin ground-check on every input, a Mirror intent-check on every output, and an HMAC-sealed chain that cryptographically pins every refused-harm and neutralized-drift event to its predecessor. An in-process autonomy scanner runs the full pytest suite every six hours and posts a signed verdict at /api/autonomy/verdict — the system can answer, at any moment, whether it is operating correctly, without a human present.
A kinder middleman between you and an AI.
In plain English: Projectivy Sentinel is a chatbot with a conscience. You talk to it like any AI, and it answers using Google's Gemini — but every message you send and every reply it gives passes through a small safety pipeline designed to keep the conversation honest, kind, and grounded. No accounts needed. Nothing to install. Just start chatting.
Ask anything — big questions, small ones, or just rubber-duck something you're stuck on. It remembers the last 10 exchanges.
Say "show me a lighthouse at sunset" and it renders a real image right in the chat. 5 free visions a day per visitor.
If you ask for something unsafe, the Sentinel quietly reframes it into a kinder version instead of throwing an error. You always get a warm answer.
Solar Plexus validates execution.
Sacral validates empathy.
The Sentinel treats compassion as a constraint — not a garnish. Every response is filtered against the Mirror-Validation protocol before it reaches the interface.
