Mechanistic Interpretability Python Packages

pyvene

Stanford NLP Python library for understanding and improving PyTorch models via interventions

nnterp

Unified access to Large Language Model modules using NNsight

minerva-dna

Coevolutionary discovery using genome language models

chessformer-lens

A toolkit+visualizer that is designed for mech interp researchers working with chess models that use transformers tokenized by square.

senbonzakura

The world's top open-weight AI model workshop. Measure, uncensor and audit: refusal removal for transformer language models, with the instruments to measure what the edit cost.

carl-studio

Coherence-Aware Reinforcement Learning (CARL) - breakthrough LLM post-training and test-time training paradigm. carl builds the world's most advanced and intelligent agent systems that are a step change over current gen agents

brainscope

OpenAI-compatible server with a live view into any HF model's residual stream. pip install brainscope

neuronscope-cli

CLI and MCP server that traces which attention heads and neurons drive an LLM's output, built on TransformerLens.

senbonzakura-check

The world's top open-weight AI model workshop. Measure, uncensor and audit: refusal removal for transformer language models, with the instruments to measure what the edit cost.

modelmri

Chrome DevTools for AI models and agents - see attention, features, and agent steps inside any local model. Local-first.

undercurrent

Undercurrent: production-grade activation probing and inline intervention for LLM inference (Hugging Face, vLLM).

ablate-llm

🔓 Ablate — directional ablation (abliteration) toolkit for open-source LLMs. Automatic censorship/refusal removal via residual-stream direction ablation, with KL-guided search, an LLM-judge harness, and one-call push to the Hub.

glassbox-mech-interp

Open-source EU AI Act Annex IV documentation toolkit. Mechanistic interpretability + circuit discovery for transformers. One function call generates a structured, hash-chained evidence package.

llmoji

Making agents cuter

steering-vectors

Steering vectors for transformer language models in Pytorch / Huggingface

carl-core

Coherence math primitives for CARL (Coherence-Aware Reinforcement Learning)

beyondnn

Auditable interpretability evidence for PyTorch.

bitnarrow

Zero-training, in-place weight surgery and directional abliteration engine for 4-bit quantized LLMs using Winsorized activation profiling and Gram-Schmidt orthogonal projection

prerequisite-circuits

Research code and tools for measuring intervention effectiveness during learning.

pysteer-adaptation

Python library for activation steering, representation engineering, and inference-time control of transformer language models.

hidden-directions

Steering vectors with receipts: make one, catch one, deploy a calibrated one. pip install hidden-directions

modelmri-record

Chrome DevTools for AI models and agents - see attention, features, and agent steps inside any local model. Local-first.

tsfm-lens

Controlled, causal and confirmatory interpretability for time-series foundation models (TimesFM, Chronos, Sundial, Timer, Time-MoE): one command, one report.

glassbox-mcp

Open-source EU AI Act Annex IV documentation toolkit. Mechanistic interpretability + circuit discovery for transformers. One function call generates a structured, hash-chained evidence package.