NousResearch/neural-steering

Mapping out the "memory" of neural nets with data attribution

A library for mechanistic interpretability of GPT-style language models

Steer LLM outputs towards a certain topic/subject and enhance response capabilities using activation engineering by adding steering vectors

The Steerability toolkit is an extensible library for general purpose steering of language models.

Steering-first neuron explorer for Qwen3.5-4B: push every MLP neuron, keep every rollout, browse the results

Extract residual-stream activations and apply steering vectors (including activation oracles) to any vLLM model during inference.

Companion code for the global workspace interpretability paper