An agent is just a loop: give a model a goal, let it read some data, check the result against your rules, and repeat until the job is done. With the Jenvu API you can build a small agent that prepares a daily XAU/USD summary for you.
1. Define the job clearly
Write one short instruction that says exactly what the agent should produce. For example: "Read today's notes, list the key levels, the session bias and the risk events. Keep it under 150 words. Never promise a result."
2. Give it the data
Pass your own inputs in the messages: price levels you track, session times, a calendar of news events. The model only knows what you send it, so keep the inputs clean and dated.
3. Check the output
Before you show or save the summary, run simple checks in your code:
- Is every level a real number from your input?
- Is the length within your limit?
- Does it avoid words like "guaranteed" or "sure win"?
If a check fails, send the reply back to the model with a short note about what to fix.
4. Pick the right model per step
You can use a fast model such as gpt-5.5 for drafting and a deeper model such as claude-opus-5 for a final review. Every model costs the same $3 per million tokens through one Jenvu key, so you can choose by quality and speed alone.
5. Keep a human in the loop
An agent should prepare information, not place trades on its own. Let it alert you, then decide yourself. Gold can move fast and leverage magnifies losses as well as gains.