Spike: server-side inference proxying behind Calypso (WAF-style deterrence)

[f5-aryan]

Proposal

Route inference through the Calypso servers so guardrail violations are blocked server-side and offending data is never sent to the client.

Rationale

Stronger Threat Deterrence Model — the WAF-at-the-server side of the analogy in #2. Prevents exposure rather than detecting it after the fact. Addresses the root cause behind #1 and #2.

Scope

• Feasibility of positioning Calypso as the inference gateway/proxy with OpenAI/Codex.
• Integration effort and trade-offs: latency, feature parity, GUI compatibility.
• Recommendation on whether to position server-side protection as the primary model with the customer.

Related

• Stop hook can't suppress responses in the Codex GUI (scan runs after display) #1 — Stop hook can't suppress responses in the Codex GUI
• Client-side hook model is bypassable and exposes data at the endpoint #2 — Client-side hook model is bypassable and exposes data at the endpoint

Raised by a collaborator during Codex hook integration review.