StartLatent docs

How Latent scores each answer your model writes for hallucinations, how it learns your traffic, and where it runs.

Latent labels every answer your model gives, in milliseconds and with no human labelers, so you can post-train it on your own traffic. On hosted plans, Latent's own reader model reads each finished answer. When Latent runs in your environment beside an open-weight model on vLLM, it reads your model as it writes. To get started, start free, or see the plans.

Start in two minutes

Install the SDK, set two environment variables and add one line at startup. Every OpenAI, Anthropic and Google GenAI call in the process is scored from then on.

pip install runlatent
export LATENT_API_URL=https://app.runlatent.ai
export LATENT_API_KEY=lk_...

# then, once at the start of your app
import runlatent
runlatent.auto_instrument()

How Latent works

Integrations

Where Latent runs

Was this page helpful?