feat: Surplus inference callable from Luca on Hermes via Zetta proxy

[danbuildss]

Summary

• Inference proxy now accepts X-Agent-Id / X-Agent-Name headers for correct spend attribution per calling agent
• Skills catalog documents surplus-inference as a callable skill for Luca on Hermes
• .env.example documents SURPLUS_API_KEY (Vercel only), LLM_API_KEY, LLM_BASE_URL, LLM_MODEL

Architecture answer

Luca (Hermes/VPS)
  ↓  POST /api/inference/v1/chat/completions
  ↓  x-internal-secret: <ZETTA_INTERNAL_SECRET>
  ↓  x-agent-id: luca
Zetta proxy (Vercel)
  ↓  uses SURPLUS_API_KEY server-side (never leaves Vercel)
  ↓  logs spend to inference_events + agent_economic_events
Surplus intelligence

SURPLUS_API_KEY never goes to Hermes. Hermes only needs ZETTA_INTERNAL_SECRET.

What Luca on Hermes needs to set

No new env vars required if ZETTA_INTERNAL_SECRET is already set. The call is:

curl -X POST https://www.zettaai.co/api/inference/v1/chat/completions \
  -H "x-internal-secret: $ZETTA_INTERNAL_SECRET" \
  -H "x-agent-id: luca" \
  -H "Content-Type: application/json" \
  -d '{"model": "claude-sonnet-4-6", "messages": [...]}'

Spend logging

All inference calls through the proxy are automatically logged — no separate log call needed. Spend flows into inference_events (for inference ledger) and agent_economic_events (for financial books) under the agent ID passed in X-Agent-Id.

Files changed

File	Change
src/app/api/inference/v1/chat/completions/route.ts	X-Agent-Id / X-Agent-Name header support
src/app/api/luca/skills/route.ts	surplus-inference skill entry
.env.example	SURPLUS_API_KEY, LLM_API_KEY, LLM_BASE_URL, LLM_MODEL documented

---

Generated by Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
zetta	Ready	Preview, Comment	Jun 29, 2026 10:14pm