pvperf in ChatGPT
ChatGPT runs pvperf through a GPT with an action. The action calls pvperf's REST API with your API key.
You need a pvperf API key. Sign in to the workbench with your email, open the Account page and create a key. The key starts with pvp_live_.
A GPT with a pvperf action
- In ChatGPT, open GPTs and click Create. Open the Configure tab.
- Enter the name
pvperfand paste the instructions below. - Under Actions, click Create new action.
- Under Authentication, choose API Key and the auth type Bearer. Paste your key and save.
- Under Schema, click Import from URL and enter
https://pvperf.com/v1/openapi-actions.json. - Click Create and choose Only me.
The instructions for the GPT:
Use the pvperf action to simulate PV systems. For a location and a system size, call
simulate_simple. To change anything else, edit the `request` of its answer and send it to
simulate_pv_system. Look up a field with docs_page, for example array.gcr. Report energy in kWh,
specific yield in kWh/kWp and losses with the unit of their loss-diagram row.
The action has four operations.
| Operation | What it does | Charged |
|---|---|---|
simulate_simple |
Runs a system from a location and a size. Returns the full request it ran. | Yes |
simulate_pv_system |
Runs a full request, for example the edited request of a simple run. | Yes |
get_run |
Reads one of your stored runs. | No |
docs_page |
Returns the documentation page of a field, a method or a loss row. | No |
A GPT shared with other people runs on your key. Your account pays for their runs.
OpenAI retires GPTs on 11 December 2026. Its retirement FAQ gives the dates for each plan.
ChatGPT plugins with an MCP server
Not live: ChatGPT connects to an MCP server only with OAuth or without authentication. pvperf's MCP server takes an API key. Use a GPT with a pvperf action instead.
A first prompt
Use pvperf to simulate 16 modules of 400 W flush on a south-facing roof at 25° in Springfield, Illinois (latitude 39.80, longitude −89.65). Give the energy to the grid for the year and for each month, and the three largest energy losses.
ChatGPT asks to allow the call to pvperf.com and calls simulate_simple once. The answer contains:
- the DC nameplate, 6.4 kWp
- the energy to the grid for a typical year, in kWh, and for each month
- the specific yield in kWh/kWp and the performance ratio
- the loss diagram, with irradiance rows in kWh/m² and energy rows in kWh/kWp
defaults_applied, each value the run assumed with its source, such as the inverter size and the typical yearbilling, the run's units and price
The run is stored with your account. get_run reads it again by its run_id.