Market ·

Jev: fast and accurate decisions in our agent engine

BrainChat agents now use Jev, a model that doesn't write text: it decides the next step based on probability.

By Lucas Marchon, Founder at BrainChat

Jev, by TypeSafe, arrived in early access in September 2026 as a new type of AI model. It doesn't generate text. It receives the conversation state and objective questions, returning typed decisions: a choice between options, a score, or a yes/no, each with its associated confidence.

How we use it

Since September 24, 2026, Jev has been running in the BrainChat agent engine alongside the model that writes the responses:

  • Turn focus: decides whether the agent should consult the knowledge base, search a table, collect a piece of data, or conclude.
  • Action judge: checks, before running an action (scheduling, recording, charging), whether it makes sense at that moment, following the agent's instructions.
  • Flow triage: when the customer replies in their own words instead of tapping a button, Jev chooses the right option.

What you get

Fewer actions triggered at the wrong time, more focused responses, and triage that understands the customer, without slowing down support.

Source: TypeSafe documentation.

Related

Explore the feature: AI agents · Flow

See all posts