Edit the model. Compose interventions. Export steered checkpoints.
Democratizes the "edit-the-model" capability that today lives in five labs. Drag features in. Suppress one, amplify another. Preview output live. Ship the recipe.
Drag-and-drop steering. Compose multi-feature interventions. Live output preview. Save as recipe URL.
Public marketplace of steering packs — helpful, honest, harmless. Fork, star, PR. One-click plug-in to your transformers model.
Upload a failure dataset, get back the top 50 features correlated with errors. The sacred grail of LLM debugging.
"What if this token were X?" Surgical replay with exact-token edits; see how features downstream shift. Feels like a React time-travel debugger.
Labs, safety teams, and individual researchers using SAE interventions in their workflow get priority access to Sandbox beta, free credits, and direct influence on the Recipe Store schema. We want your hardest use case.