Everything you need to navigate the space between voices.
DRIFT runs entirely in the cloud — no local installation required. Just sign up and open the app in your browser.
Sign in, navigate to your coordinates, enter an idea, and hit Generate. Your first generation may take a minute while the model warms up — subsequent generations are faster.
Each knob positions the language model along a specific dimension. Turning the knob doesn't change the topic — it changes the voice that speaks.
Shifts the epistemological worldview. At one extreme the model speaks from inside an oral tradition where revelation is evidence. At the other, from a fragmented future where the question itself may be obsolete.
Controls how the model speaks, not what it says. The same idea expressed in raw street language versus dense academic prose. Combine with Time for compounded effects.
Moves the model between open inquiry and absolute declaration. Low certainty produces lyrics full of questions and possibility. High certainty produces proclamation.
Pulls the output toward or away from conventional coherence. At high values the model begins to fragment, associate freely, and produce genuinely strange language.
Select up to 3 genres from 6 families. Selecting multiple genres creates a fusion — the style prompt combines them. Arrows navigate between genre families.
A patch saves your current coordinate position — all four axis values — under a name. Load a patch to return to that exact position instantly. Useful for building a vocabulary of voices you return to.
Physical control surface integration for the Stream Deck+ is in development. Four dials will map directly to the four axes, with the touch strip handling genre navigation. No Elgato software required — the bridge talks directly to the hardware.
Available on Pro plans and above. Authenticate with your API key from the dashboard.