Silicon Psyche Labs · Layers

The layers we put on a model

From the outside in: what reads the model's words, what watches a swarm of agents, what reads the model's internal state, and what we change in its weights. Pick a scenario to watch which layer acts, or click any layer.

How we fine-tune a model, step by step →
Live in the product todayMeasured in the lab, on models we host
Moving dots illustrate the flow.
Scenario
What acts, in order
    Readings in the steps illustrate what each layer returns. Measured numbers are in the layer card.
    Not a gate at the door

    How this differs from a guardrail

    In plain words