toolspool
Aquin logo

Aquin

verifiedFreeaquin.app

Mechanistic-interpretability tooling to debug and improve ML models-trace outputs, simulate fine-tunes and patch failures.

What it does

Aquin (Aquin Labs) builds mechanistic-interpretability tooling for debugging and improving ML models. Using sparse autoencoders, activation patching and causal tracing, it lets engineers observe features and layers, trace an output token to the prompt span that caused it, and simulate training runs to catch failures. A CLI plugs into existing training stacks and mirrors to a web panel.

How to use: Use the command bar for AI search, interact with webpages using Ctrl+F for chat mode, upload files for AI insights, and create custom shortcuts for specific AI actions. Utilize Zen Mode for distraction-free browsing.

Core features

Interpretability tools (SAE, logit lens, causal tracing)
Token-to-prompt/layer attribution
Simulate LoRA/QLoRA/DPO/distillation runs
Diff base vs fine-tuned models
Aquin CLI for existing training stacks
Supports dense LLMs, MoE, vision and embeddings

Best for

→Debugging model hallucinations and drift
→Understanding which circuits produce outputs
→Testing fixes before full fine-tuning
→Auditing model behavior and safety
Toolspool rankingby monthly traffic

Tutorials

Step-by-step: exactly how to get things done with it.

Aquin is out. Vibe code your own LLMs in few minutes!
Aquin · 14K views
Aquin — AI Browsers | Toolspool