fix: v1Auth surplus inference endpoint for Luca on Hermes Telegram runtime

[danbuildss]

Root cause

The previous inference endpoint (/api/inference/v1/chat/completions) used internalAuth — it requires ZETTA_INTERNAL_SECRET. Luca on Hermes uses a v1 API key for all skill calls, not the internal secret. Auth mismatch = 401 on every call from Telegram runtime.

Fix

New POST /api/v1/luca/inference — same auth as every other Luca skill (Authorization: Bearer <api-key>). No new env vars needed on Hermes.

• OpenAI-compatible body: { model, messages, stream? }
• SURPLUS_API_KEY stays on Vercel, never sent from Hermes
• Spend logged automatically to inference_events + agent_economic_events
• Skills catalog updated to point to the correct endpoint with correct auth docs

Test from your Mac (verify Vercel is working before telling Luca)

curl -X POST https://www.zettaai.co/api/v1/luca/inference \
  -H "Authorization: Bearer <LUCA_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{"model":"llama-3.3-70b","messages":[{"role":"user","content":"reply with one word: working"}]}'

Should return an OpenAI-compatible response. Check inference_events in Supabase — a row with provider=surplus, agent_id=luca should appear.

What to tell Luca on Hermes

"New endpoint for Surplus inference: POST https://www.zettaai.co/api/v1/luca/inference — use your existing API key (same one you use for agent-read, agent-books, etc.). Body: { model, messages }. No other changes needed."

---

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 10:30pm