This is a fresh take on LLM agents. Most frameworks rely on loose ReAct loops, so a deterministic OS-style pipeline is an interesting approach. Looking forward to exploring the repo.

Thank you! That’s exactly the reaction I was hoping for.

Most agent frameworks treat the LLM like a “smart intern” — I wanted to treat it like a superintelligence that needs to be kept in a constitutional prison with multiple immune systems

The entire pipeline is deliberately rigid and deterministic on purpose: every single decision has to go through Evidence → Critique → Debate → ValueCore → FUJI Gate → hash-chained TrustLog. No shortcuts, no “hope the model behaves”.

Would love to hear your thoughts once you dig in — especially on whether the safety overhead feels worth it in practice.

(Also happy to answer any questions about the Doctor Dashboard or the Telos two-axis value system!)