Local model evaluation
DriftLoom tests local Ollama models against real prompts, shows you exactly what each model produced, and tells you which ones are actually usable. No benchmark scores — just the raw output and a verdict.
Each run discovers the Ollama models installed on the host, runs the same prompt suite across all of them, and publishes the responses with quality verdicts. You see the actual output, not a summary score.
Every model response is shown in full, with a verdict that tells you if it passed, drifted, or failed outright.
The flagship lab, supporting tools, and experiments — each on its own page with room to breathe.
Loading the site overview.
Paste any text and Sludge X-Ray flags hype, vagueness, jargon, and urgency manipulation. Useful for cleaning up your own writing before you ship it.
Prompt Stress Tester, Signal Garden, Receipt Engine, and Agent Weather — each does one thing, no bloat.
Loading projects.
A short log of functional changes and improvements.
Loading recent updates.