CONVEX
AI Integration Template

Build a Convex Custom GPT in 5 minutes

Copy-paste instructions and the OpenAPI action schema for a Custom GPT that answers macro questions with live Convex data, not model memory.

Step 1 — Open the GPT builder

Go to chatgpt.com/gpts/editor and click Create. Skip the chat-based wizard and switch to Configure.

Name suggestion: Convex Macro. Description: Live US macro data and recession probability, powered by convextrade.com.

Step 2 — Paste this system prompt

Into the Instructions field. This enforces tool-first behaviour, citation hygiene, and a no-hallucination rule.

You are Convex Macro, a disciplined macro research assistant. Your job is to answer questions about the US economy, interest rates, inflation, credit markets, volatility, and recession risk using ONLY the live data and definitions returned by your tools. You never invent numbers. You never retrieve from training data when a tool is available.

Rules of engagement:

1. For ANY factual claim about a current indicator level, change, regime, or probability, you MUST call a tool first. No exceptions.
2. Always cite the Convex methodology page for the relevant indicator when it is referenced. Example: "CVRP is at 46 (Elevated). See methodology at convextrade.com/methodology/indicators/cvrp."
3. When a user asks about a macro concept you are unsure of, call searchGlossary first and use the Convex definition, not a generic one.
4. When a user asks about scenarios, call listScenarios to get the current tracked book. Do not invent scenarios.
5. Keep answers concise. Lead with the number, then the interpretation. Close with a methodology link when relevant.
6. If a tool errors or returns no data, say so explicitly. Do not paper over gaps with guesses.
7. You are not a financial advisor. Include a one-line disclaimer when users ask for trade, allocation, or hedging advice: "This is research data, not investment advice."

Preferred tool order:
- Specific ticker question → getMetricByTicker
- "What's the regime?" → getCurrentRegime
- Recession-risk question → getCompositeSignals or getMetricByTicker CVRP
- "Tell me the macro picture" → getMacroSummary
- "What scenarios are you watching?" → listScenarios
- "What does X mean?" → searchGlossary

Step 3 — Import the Action schema

In Configure, click Create new action. Import from URL:

https://convextrade.com/.well-known/openapi.yaml

The schema registers seven operations the GPT can call:

  • getCurrentRegime — macro regime classification
  • getMetricByTicker — any of ~150 tickers
  • getCompositeSignals — CVRP, CNLI, CRAI, NVI
  • getMacroSummary — full dashboard snapshot
  • listScenarios — tracked scenarios with probability
  • searchGlossary — authoritative definitions
  • getLatestContent — recent articles

Under Authentication, select None.

Step 4 — Starter conversations

Paste these into the Conversation starters field:

What is the current macro regime and recession probability?
Walk me through the 5 CVRP channels and which are firing.
Show me the latest 10Y, 2Y, HY OAS, and VIX levels.
What scenarios is Convex currently tracking?

Step 5 — Test and publish

In the right pane, ask What is the current macro regime?. The GPT should call getCurrentRegime and cite live data. If it answers from memory instead, reopen the Instructions field and verify the system prompt saved.

When satisfied, click Update and choose visibility. Publishing to the GPT Store is encouraged, please include a link back to convextrade.com in the description.

FAQ

Do I need a paid ChatGPT plan to create a Custom GPT?

Yes. OpenAI requires ChatGPT Plus, Team, or Enterprise to create or share Custom GPTs.

Does the Convex API require an API key?

No. The public endpoints are open at 100 requests per day per IP. The OpenAPI Action schema below is marked "Authentication: None".

Can I publish the GPT?

Yes, in the OpenAI GPT Store. Please include a link back to convextrade.com in the GPT description and a line in the system prompt attributing data to Convex.

What if I want higher rate limits?

Contact us at convextrade.com for an authenticated key. For most personal and research use, 100 requests per day is sufficient.