Synecdochesin·EK·duh·kee · The part for the whole. The whole for the part.
AI for the people.
A personal AI that lives in your texts. It remembers the thing you said you'd do, keeps the list, finds the flight, and sits in the family group chat when you add it. Name it anything. Your data is yours: not sold, not used to train, gone when you say.
Text (646) 288-1416Free while we build it. Standard rates. Reply STOP any time.
What it costs to run, in electricity.
No one is required to tell you what an AI answer costs in power. We think you should know, so here is ours, in things you can picture.
2 answers for 1 person since September 28, 2026, not counting our own testing. One answer keeps the bulb lit about 2 minutes (2.4 Wh).
Our goal, over time, is to keep bringing this number down, and to make Synecdoche a leader in clean, energy-efficient AI. You can ask yours for its own number any time: “how much electricity have you used?”
How we count
No model vendor publishes energy per token, so this is our estimate. We count every token we send and receive for every answer, then scale from the measurements that exist: Google's 0.24 Wh for a median Gemini text prompt (August 2025, full stack including idle capacity and data-center overhead) and Epoch AI's 0.3 Wh for a typical ChatGPT query (February 2025). Generating a token costs the most, reading one far less, reading one from cache less again: 0.0006, 0.00012 and 0.000012 Wh each. A phone charge is counted as 15 Wh, a kettle boiled as 100 Wh, a hair dryer as 25 Wh a minute. Our answers carry more context than a chatbot's, so our number per answer is higher than theirs; that is the honest figure, not a flattering one. The numbers update every few minutes and are totals only, never any one person's.