Choose
Select the best action from a constrained set of possibilities.
Krun is building machine-native models for decisions inside software, agents, and autonomous systems.
Customer support task with account context and refund policy.
Designed for software that needs structured outputs, explicit uncertainty, low latency, and predictable behavior.
Select the best action from a constrained set of possibilities.
Order alternatives by relevance, quality, risk, or expected outcome.
Estimate intent, value, quality, risk, or any measurable decision surface.
Choose the right model, agent, tool, or workflow for the current state.
Decide whether an action should execute, escalate, or require review.
Typed outputs your software can act on directly — with confidence your policies can use.
{
"decision": "approve",
"confidence": 0.91,
"action": "execute"
}Set operational thresholds for autonomy, escalation, or human review.
Modern language models are extraordinary at generating language. Autonomous software requires something different: predictable outputs, explicit uncertainty, low latency, low cost, and decisions that compose cleanly in code.
Krun is exploring models designed around those requirements from the beginning.
Krun-1 is on its way. Private preview, API access, and research releases are coming soon.