How to Use Renewable-Powered AI on magicdoor.ai
Magicdoor lets you choose renewable-powered inference for three supported open-weight models. The Renewable control changes the inference route for supported-model requests while it is selected in chat; it does not claim that the models were trained using renewable energy.
Choose the model. Choose renewable.
Supported models
| Model in Magicdoor | GreenPT model route | Attachments on renewable route |
|---|---|---|
| GLM-5.2 | glm-5.2 | No |
| Kimi K3 | kimi-k3 | Images only |
| DeepSeek V4 Flash 0731 | deepseek-v4-flash-0731 | No |
PDFs and other documents are not supported on this route. GreenPT lists these model IDs in its current model API documentation.
Turn it on in chat
- Open Magicdoor chat.
- Select GLM-5.2, Kimi K3, or DeepSeek V4 Flash 0731.
- Press the Renewable toggle beside the model selector.
- Send your prompt as usual.
- After the reply, expand Renewable inference beneath the response to inspect the receipt.
The Renewable toggle appears only when the selected model supports this route. If you switch to an unsupported model, that model continues on its existing inference route.
Prefer renewable inference by default
Only active subscribers can save Prefer renewable inference in chat preferences. Saving it applies the choice to the current chat and loads it when new chats start. Magicdoor will use GreenPT whenever the active model supports it.
Without an active subscription, select a supported model and turn on Renewable beside the model selector. While the control is on, supported-model requests use GreenPT; starting a new chat reloads the saved preference.
“Prefer” is deliberate: it does not reroute every model. Unsupported models keep their normal behavior, and Magicdoor tools remain available. The GreenPT route itself keeps Magicdoor tools available for supported requests.
What stays unchanged
- Your prompt-first chat workflow
- The ability to switch models during a conversation
- Magicdoor tools
- The existing routes for unsupported models
- The explicit choice of model—you still choose what answers
The route is attributed to GreenPT, which says the inference runs on 100% renewable-powered infrastructure in the EU. This is an inference claim, not a training claim, data-residency promise for Magicdoor as a whole, or full-platform certification claim.
Read the response receipt
After a renewable response, expand Renewable inference. Magicdoor displays provider-reported values for:
- Inference time: how long the model computation took, in the provider's reported unit
- Energy: an estimate normalized for display in mWh or Wh when possible
- Emissions: estimated CO2e for the inference operation
The receipt also offers familiar context, such as time for a 10 W LED bulb. Its method is marked alpha. Comparisons provide scale; they are not direct comparisons with Magicdoor's other inference route. The impact object is transient and session-only, so it will not remain as part of the stored chat after the browser session.
For the units and calculation boundaries, read AI energy use and carbon footprint: what one answer can measure. For the broader terminology, read what green AI means.
Attachments and troubleshooting
The Renewable toggle is missing. Check that you selected one of the three supported models. It is intentionally hidden for other models.
An attachment is rejected. On the renewable route, only Kimi K3 accepts image attachments. Remove the attachment, choose Kimi K3 for an image, or turn off Renewable and use a compatible existing route. PDFs and documents are not supported.
There is no receipt. Confirm Renewable was active when you sent the prompt. A provider impact object must be returned for Magicdoor to show the receipt.
A tool is needed. Magicdoor tools remain available on the GreenPT route. Enabling the preference does not disable them.
Related Resources
What Is Green AI? Renewable-Powered AI Explained
A practical guide to sustainable AI: renewable inference, efficient models, data-center PUE and WUE, offsets, training, and transparent measurement.
GPT-5.6 Luna Guide - Efficient OpenAI Work on magicdoor.ai
A practical guide to GPT-5.6 Luna pricing, supported tools, everyday use cases, and when to switch to GPT-5.6 Sol.
AI Energy Use and Carbon Footprint: What One Answer Can Measure
Understand Wh, mWh, CO2e, inference time, PUE, WUE, and the limits of per-answer AI energy and emissions estimates.
AI Cost Optimization: A Practical Model Routing and Budget Guide
A decision framework for controlling AI costs with current chat and image models, measured usage, model escalation rules, and an honest flat-rate break-even check.