Documentation

Everything you need to use Drift.

Setup Guide

Drift runs the language model locally on your GPU. You need Python 3.12 and an NVIDIA GPU with at least 8GB VRAM.

Requirements

Python 3.12 — python.org/downloads
NVIDIA GPU with 8GB+ VRAM (RTX 3060 or better recommended)
Windows 10/11 or Linux
Suno API key — sunoapi.org/api-key
~15GB free disk space (for model weights)

Install dependencies

py -3.12 -m pip install flask python-dotenv requests transformers torch --index-url https://download.pytorch.org/whl/cu124 repeng scikit-learn accelerate

Configure your API key

Create a file at ~/.env with your Suno key:

SUNO_API_KEY=your_key_here

Run the server

cd drift
py -3.12 server.py

The first run downloads Mistral-7B (~10GB) and trains the time vector (~10 minutes). Subsequent runs load from cache in about 30 seconds.

The Four Axes

Each axis moves the model along a different dimension. Values range from -2.0 to +2.0 (0 to 1.0 for Weirdness). Double-click any knob to reset it to center.

Axis	Range	Negative pole	Positive pole
Time	-2.0 to +2.0	2050 post-AGI	7th century Arabia
Register	-2.0 to +2.0	Plain speech	Scholarly elevation
Certainty	-2.0 to +2.0	Exploratory doubt	Declarative certainty
Weirdness	0 to 1.0	Coherent	Unhinged

API Reference

Pro and above plans have API access. All requests require a Bearer token from your dashboard.

POST /generate

POST https://api.godrift.fm/generate
Authorization: Bearer YOUR_API_KEY
Content-Type: application/json

{
  "text": "what to write about",
  "state": {
    "time": 1.8,
    "register": 0.8,
    "certainty": 0.6,
    "weirdness": 0.4,
    "voice": "both"
  },
  "count": 1,
  "genres": ["qawwali", "doom"],
  "tags": ["cinematic"]
}