hijohnnylin/neuronpedia

Health v2 · maintenance only

Adoption not part of the score

1124 stars · 147 forks observed · 2026-09-03

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Neuronpedia is an open source AI interpretability platform for exploring, visualizing, and steering the internals of language models. It hosts terabytes of sparse autoencoder (SAE) feature data, activation dashboards, circuit tracing, autointerp explanations, and a JSON API, both as a hosted site (neuronpedia.org) and a self-hostable webapp.

Use cases

• browse sparse autoencoder features of language models
• search which SAE features activate on my text
• visualize neuron activations and top logits for a feature
• embed an interpretability feature dashboard in my site or paper
• steer an LLM's behavior using feature vectors
• trace a model's internal reasoning with circuit graphs
• find misaligned persona features in open-weight models
• query feature data programmatically via a JSON API

When to choose

• you do mechanistic interpretability research on LLMs and need feature/SAE dashboards
• you want a hosted or self-hosted browsable database of activations and autointerp explanations
• you need embeddable visualizations of model internals for papers or demos
• you want to test features on custom text or steer models via an API

When to avoid

• you need general-purpose ML training or fine-tuning tooling
• you want a lightweight library rather than a full web platform with a database
• your work doesn't involve inspecting neural network internals
• you need offline tooling without running inference infrastructure

Facets

application · maturity active

machine-learning data-visualization search-engine api-framework llm-inference rag artificial-intelligence machine-learning large-language-models data-visualization developer-tools self-hosted interpretability mechanistic-interpretability sparse-autoencoders sae neuron-visualization feature-dashboards activation-analysis circuit-tracing autointerp steering llm-internals open-source-platform web-server nodejs typescript docker

Member repositories

For agents

markdown · JSON · MCP: product_card(name="hijohnnylin/neuronpedia")